science.magazine.5684.2004-07-30

www.sciencemag.org SCIENCE VOL 305 30 JULY 2004 569

Strained and Stretched Nanoparticles The electronic and optical properties of a material can changeon going from bulk materials to the nanoscale. Gilbert et al.(p. 651, published online 1 July 2004) show how confinementeffects can affect the bonding and packing of atoms. They use anumber of techniques to measure the lattice structure and in-ternal strains in 3-nanometer particles of zinc sulfide. A com-plex pattern of internal strains results as the particles attemptto lower the surface energy. These strains cause a reduction inthe overall ordering and a stiffening of the lattice. When metalsare deformed, the crystallinegrains can rotate and realign,much in the way that paintedshapes will stretch and warpwhen a canvas is pulled in aspecific direction. Shan et al.(p. 654; see the Perspective byMa) find that in nanocrystallinenickel, this type of deformationdominates, unlike the situationwith coarser grained metals,where the production of grainboundary defects and disloca-tions accommodates most ofthe deformation energy. Theseresults confirm many observa-tions obtained from computersimulations and should helpguide the design of optimummetals and alloys.

Lunar MeteoritePhones HomeA lunar meteorite found in theSultanate of Oman (Sayh alUhaymir 169) consists of fourdifferent impact breccias and isenriched in potassium, rare earth elements, and phosphorus.Gnos et al. (p. 657; see the Perspective by Korotev) used iso-topic systematics to date the four impact events that occurredwhile the rock was at or near the surface of the Moon. The im-pact event dates of 3900 million years ago (Ma), 2800 Ma, 200Ma, and <0.34 Ma, along with the chemical enrichments, helpto pinpoint the source of the meteorite in the Lalande impactcrater on the Moon. The dated impact events will allow lunargeologists to refine the ages of the different stratigraphic unitsassociated with this meteorite into a more global model of theevolution of the Moon.

Removing Plant Defenses In order to resist herbivore attack, plants use direct defenses,such as toxins and digestibility reducers, as well as indirect de-fenses that affect components of the plants’ community (suchas natural enemies and diseases). Plant defenses can be ex-pressed constitutively or produced in response to an attackingpathogen or herbivore. Kessler et al. (p. 665, published online 1July 2004; see the Perspective by Dicke et al.) transformed thewild tobacco species Nicotiana attenuata, to silence three

genes coding for enzymes in the jasmonate signaling pathway,which is known to be involved in induced plant defense. Whenplanted into native habitats, the transformed plants were morevulnerable not only to their specialist herbivores but also toother herbivore species.

Pop Goes the MitochondrionIn cells undergoing apoptosis or cell death, mitochondria, thepowerhouses of the cell, often have a key role. Not only is cellu-

lar metabolism shut down, butmitochondria release moleculesinto the cytoplasm that furtherpromote cell death. Substantialcontroversy has surrounded themechanisms by which theseprocesses occur. Green and Kroe-mer (p. 626) review the role ofmitochondria in cell death. Per-meabilization of the mitochondriacan be the point-of-no-returnthat seals the fate of a cell, andnumerous strategies are envi-sioned to alter these processestherapeutically to benefit patientssuffering from a range of illnessesfrom cancer and heart failure toneurodegeneration.

Herbivores Drive DiversityHabitat specialization and beta-diversity—the change in speciescomposition between sites—mayexplain a large part of the overalldiversity within tropical forests.However, why beta-diversity

should be higher in the tropics remains unclear. To test the hypothesis that herbivores promote habitat specialization, Fineet al. (p. 663; see the Perspective by Marquis) performed recip-rocal transplant experiments of specialist tree seedlings be-tween soil types in the Peruvian Amazon, and also manipulatedtheir herbivores. Habitat specialization of plants resulted froman interaction of herbivore pressure with soil type, which sug-gests that herbivores drive beta-diversity patterns by maintain-ing habitat specialization.

Diffusion Goes Electronic The atomic-scale resolution of the scan-ning tunneling microscope has been pairedwith the temporal resolution afforded byfemtosecond laser pulses to differenti-ate electronically excited moleculardiffusion from thermally induceddiffusion. Bartels et al. (p. 648,published online 24 June 2004) excit-ed CO molecules on the anisotropic Cu(110)

edited by Stella Hurtley and Phil Szuromi

THIS WEEK IN

CONTINUED ON PAGE 571CR

ED

ITS:(

TO

P T

O B

OT

TO

M) YA

MA

DA

ET

AL.

;BA

RT

ELS

ET

AL.

Controlling Interface SpinThe magneto-electronic properties of heterojunctionstructures formed between magnetic thin films and in-sulating layers are attractive for potential device appli-

cations. However, the be-havior of such structureshas been unpredictable,and techniques are need-ed that can investigatethe influence of interfaceregion. Using magnetiza-tion-induced second-har-monic generation, Yama-da et al. (p. 646) showthat they can probe andcharacterize the magne-tization of buried inter-faces formed betweenmanganite and insulatingthin films. Moreover, bygrading the doping level

of the interface region, they show how the propertiescan be altered in a controlled manner.

www.sciencemag.org SCIENCE VOL 305 30 JULY 2004

surface with 200-femtosecond laser pulses at a wavelength of 405 nanometers. Unlikediffusion at thermal or equilibrium conditions, which occurs along the rows of atomsformed by Cu atoms on this surface, the nonequilibrated electronically excited COmolecules diffused over the rows as well. A phenomenological model can account forthese results in terms of electronic excitation of the CO-substrate vibrations.

Worming into Whale BonesA new genus of annelid worm that is related to hydrothermal vent worms has beendiscovered on the corpse of a gray whale found several thousands meters deep off thecoast of California. Rouse et al. (p. 668) have named the genus Osedax. The femaleworms possess tubes from which red plumes emerge and which harbor numerous, non-feeding, dwarf male worms. Like vent worms, Osedax worms are gutless and containbacterial symbionts. The worms burrow into the whale bones and form rootlike struc-tures which contain the symbiotic organotrophic bacteria that mobilize nutrients fromthe whale skeleton.

The Genomics Underlying AcnePropionibacterium acnes is a ubiquitous, human skin–dwelling organism involved in theetiology of acne. Brüggemann et al. (p. 671) have sequenced and analyzed the com-plete genome of P. acnes. The genome data offer information on the bacterial antigensand tissue-damaging enzymes that may cause the inflammatory reactions underlyingthe disease process.

Freeze-Frames of Motor Movement Kinesin motor proteins move along microtubules by rapidly alter-nating between tightly bound and detached states. Movement is

adenosine triphosphate (ATP)–dependent and changes in bind-ing affinity are associated with the ATPase cycle. Nitta et al.(p. 678) report crystal structures of the monomeric kinesinKIF1A with three transition state analogs. Kinesin alternatelyuses two loops to bind microtubules with an intermediatestate in which neither loop binds. When KIF1A is working, itlikely alternates between a tight-binding state with the affini-ty biased toward the forward tubulin subunit, and a weak-

binding state that allows one-dimensional diffusion.

Reconstituting Prion Disease in MiceThe prion hypothesis postulates the existence of infectious proteins capable of prop-agating disease. Legname et al. (p. 673; see the news story by Couzin) now presentevidence that a novel strain of prion disease can be induced in mice injected with re-combinant prion proteins. Brain extracts from these mice could then be used to in-fect other mice to cause a neuropathological disorder distinct from other knownstrains of prion disease.

Host-Parasite Gene Transfer in Plants The parasitic plant family Rafflesiaceae resisted definitive taxonomic placementsince its initial description nearly two centuries ago. Recently, a study placed it firmlyin the order Malpighiales, based on the mitochondrial gene matR. Davis and Wur-dack (p. 676, published online 15 July 2004) have reexamined this question byadding all family representatives of Malpighiales across four genetic loci spanningthe nuclear and mitochondrial genomes. The nuclear DNA, and one mitochondrial lo-cus, confirmed the position of Rafflesiaceae within Malpighiales. However, the othermitochondrial locus, nad1B-C, places Rafflesiaceae in the family Vitaceae, which is ina different order. These incompatible phylogenetic results appear to provide a newexample of horizontal gene transfer between species—horizontal gene transfer me-diated by a plant host-parasite system.

CONTINUED FROM 569 THIS WEEK IN

CR

ED

IT:N

ITTA

ET

AL.

EDITORIAL


The Olympics are upon us, and we sports fans can’t wait for the nonstop television dietof sports molded from the classic Greek tradition, like wrestling and track, and of oth-ers added to the occasion, like rhythmic gymnastics, synchronized swimming, and evenbaseball. Political pressure in support of a primarily national sport can put it on the list,just as can the appeal of the sport itself. And the Olympics have shown little capacityto resist either the proliferation of new sports or the professionalization of old ones.

How long, one wonders, will they be able to hold out against the “extreme sports” categories nowrepresented in the X Games?

What drives this diversification is partly revenue, which of course meanstelevision. But there are other forces that have much to do with science. TheOlympics selects athletes performing at the edge of their physical capacity,pushing competitors into training regimes unheard of decades ago. Thesame urge has moved the Olympic Committees toward accepting profes-sional athletes as competitors, which surely has the sometime Olympic czarAvery Brundage revolving in his grave. This oddly sporadic surge towardprofessional acceptance yields a perplexing heterogeneity of treatment:“Dream teams” of National Basketball Association players are dispatched torepresent the U.S. and set up in luxury hotels, while America’s college base-ball players bunk with the rest of the plebeians.

Here’s another science-based change in the Games: it’s how materialsscience has transformed some of the traditional sports. I actually can re-member the first 15-foot pole vault, but after the properties of fiberglassconverted the vaulter’s instrument from a pole to a catapult, we entered anew record-setting domain. In cycling and yachting, technology probablyaccounts for more of the variance in outcome than in other Olympic sports.(A friend of mine resents this, refusing to take seriously any sport that de-pends on the device as well as the athlete.)

The big science problem, though, is that in the sports that most directlymeasure individual athletic ability, there is no guarantee that the playingfield is level. Drug violations are not new to the Games; some winning dis-tance runners were charged with blood doping decades ago, and more re-cently the Canadian sprinter Ben Johnson was stripped of his medal because of steroid abuse. Nowseveral U.S. track and field athletes, including a few prospective Olympic competitors, are undersuspicion, and others will remain radioactive until the testing regime improves enough to earn pub-lic trust. What we now have is a pharmacological arms race between the detection technology of theanti-dopers and the inventiveness of the designer-steroid mavens. It is a close contest, and if past isprologue, we cannot know who is ahead at any given moment. I liked track and field a lot more be-fore they took it into the lab.

Fortunately, the really significant performance gains have come not from drugs but from betterunderstanding of the body’s limits and the role of training in overcoming them. Dr. Roger Bannisterused elegant experiments on his own respiratory physiology to help shatter a record once thoughtunbreakable. Now dozens of runners from around the world can beat his time by 15 seconds or so.And there has been a remarkable change in our ideas about what women can do in events previous-ly dominated by men, exemplified by the rapid convergence of the women’s times in distance racestoward the best men’s times.

Science surely has had a mixed impact on the Games: It has been used to enhance human ca-pacity through improved training and better technology, but it has also brought us clever ways tocheat. As for me, even though I know that everything may not be on the level, I really am lookingforward to the Olympics. So let the Games begin! I plan to adopt the English poet Samuel TaylorColeridge’s advice and follow the events having willingly suspended disbelief, confident that theplaying field is level, that no one is on drugs, and that no athlete has a concealed bionic assist. Don’tlaugh; it works for me.

Donald KennedyEditor-in Chief

Here Come the Olympics

CR

ED

IT:R

ICK

RIC

KM

AN

/NEW

SPO

RT

/CO

RB

IS

30 JULY 2004 VOL 305 SCIENCE www.sciencemag.org586

CR

ED

IT:J

OH

N C

OG

ILL/

AP P

HO

TO

NEWSP A G E 5 8 9 5 9 1

Contentioncontinues overKennewickbones

Prion proof?Perhaps

Th is We e k

DUBLIN, IRELAND—In a public appearance thatdrew worldwide media coverage, StephenHawking claimed last week that he had solvedone of the most important problems inphysics: whether black holes destroy the infor-mation they swallow. Speaking at a confer-ence here* in a lecture hall packed with physi-cists and reporters, the University of Cam-bridge professor reversed his long-standingposition and argued that information survives.As a result, Hawking conceded the mostfamous wager in physics and handed overan encyclopedia to the winner of the bet.

“It is great to solve a problem that hasbeen troubling me for nearly 30 years,”Hawking said during his presentation.Other physicists, however, doubt thatHawking has solved the long-lived puz-zle. “It doesn’t seem to me to be convinc-ing,” says John Friedman, a physicist atthe University of Wisconsin, Milwaukee.

The question of what happens to in-formation when it falls into a black holegoes to the heart of a central idea in mod-ern physics. Just as scientists in the 19thcentury figured out that energy can beneither created nor destroyed, many 20thcentury physicists concluded that infor-mation is also conserved. If true, infor-mation conservation would be one ofthe most important principles in sci-ence—perhaps more profound eventhan conservation of mass and energy.Unfortunately, there was a big obstacle:black holes.

When an object falls into a black hole,its mass and energy leave an observableimprint by making the black hole moremassive. According to general relativity,however, any information the object car-ries is irretrievably lost: An outside ob-server couldn’t tell whether the black hole hadswallowed a ton of lead, a ton of feathers, or aton of Ford Pintos. If black holes can destroyinformation in this way, information conserva-tion cannot be a universal law.

The debate raged for decades whetherblack holes were an incurable exception to

the permanence of information. In the1970s, Hawking and some of his colleagues,including Kip Thorne of the California Insti-tute of Technology (Caltech) in Pasadena,argued that black holes trump information.Others, such as Caltech’s John Preskill, ar-gued that some undiscovered loopholewould keep information safe until the blackhole somehow disgorged it. In 1997, Hawk-ing and Thorne made a wager with Preskill;

the winner was to receive an encyclopedia ofhis choice, from which information can al-ways be retrieved.

At the Dublin conference, Hawking con-ceded the bet. Using a mathematical tech-nique known as the Euclidean path integralmethod, Hawking proved to his own satis-faction that information is not, in fact, de-stroyed when it falls into a black hole. “Ifyou jump into a black hole, your mass-

energy will be returned to our universe … ina mangled form which contains the informa-tion about what you were like, but in a statewhere it cannot be easily recognized,” saidHawking. That implies that black holes arenot portals to other universes, a possibilityHawking himself had suggested. “I’m sorryto disappoint science-fiction fans,” he said.

In conceding the bet, Hawking presentedPreskill with Total Baseball: The UltimateBaseball Encyclopedia. Thorne, however, re-fused to admit defeat. “I have chosen not toconcede because I want to see more detail,”he said, but added, “I think that Stephen isvery likely right.”

Others are less certain. Friedman, for one,has doubts about Hawking’s mathematicalmethod. Quantum field theorists are happy touse the Euclidean path integral technique forproblems involving particles and fields, butmost gravitational theorists avoid it becauseit produces equations riddled with hard-to-reconcile infinities. They prefer a morestraightforward “Lorentzian” approach togravity. Nobody has proven that the twomethods always give the same results. “I’mskeptical whether the Euclidean path inte-gral method generally represents the evolu-tion of spacetime that is really Lorentzian,”says Friedman. If not, then Hawking’s con-clusion may be an artifact of the mathemati-cal method rather than a general result. An-other reason for skepticism, Friedman says,is that Hawking’s calculation takes a sumover all possible idealized black hole loca-tions and all observers in the universe, butthe results don’t seem to apply to a specificblack hole and a specific observer.

In part because of the Euclidean method,Hawking’s work doesn’t seem to yield anyinsight into how black holes preserve or re-lease information—whether all the pent-upinformation bursts forth at once, or whetherit trickles out as subtle correlations in radia-tion coming from the black hole. EvenPreskill says he wishes that Hawking’s argu-ment made more physical sense and couldbe expressed in more conventional mathe-matical terms. “If one could extract from thecalculation an understanding that could bereproduced in a purely Lorentzian calcula-tion, that would help a lot,” he says.

Despite his doubts, Preskill has noqualms about accepting Total Baseball. “Theterms were that the winner would receive theencyclopedia when the other party con-cedes,” he says. “I don’t have to agree.”

–CHARLES SEIFE

Hawking Slays His Own Paradox,But Colleagues Are Wary

T H E O R E T I C A L P H Y S I C S

Not proven? Stephen Hawking’s new view of black

holes rests on unusual math.

* 17th annual International Conference on Gen-eral Relativity and Gravitation, 18–24 July.


5 9 4 5 9 6 6 0 0

The ultimate fluexperiment

Insights froman outburst

Neuroscienceand the deathpenalty

Foc u s

Banking on the benevolence of a lame-duckCongress is risky business. But for U.S. sci-entists, a possible postelection session maybe the best bet to salvage research programsthat are facing budget cuts.

Last week, Congress began a 6-weekbreak having finished work on only one ofthe 13 spending bills for the 2005 fiscal yearthat begins on 1 October. Although the lonecompleted bill provides modest increases fordefense research, the House has taken someinitial steps on a dozen other agencies thatsuggest most science programs are in for avery hard time this year. The National Sci-ence Foundation (NSF) and NASA are fac-ing real cuts, and the National Institutes ofHealth may have to settle for a small in-crease that may not keep pace with inflation(Science, 16 July, p. 321). The dark budgetmay hold silver linings for the Departmentof Energy’s (DOE’s) science programs andthe National Institute of Standards and Tech-nology (NIST). But even those gains couldbe at risk once Congress returns in Septem-ber for a last-gasp attempt to finish its fiscalbusiness before the November elections.

The squeeze is a result of Republican-ledefforts to reduce taxes and hold down domes-tic spending while fighting wars in Iraq andAfghanistan and defending against terrorismat home. That has left the 13 spending panelsthat divvy up the government’s $2 trillionbudget with less money than agencies re-quested. The latest bad news came on 22 July when a House panel voted to shrink thebudgets of NSF and NASA by 2.1% and1.5%, respectively, below this year’s levels.

The decline for NSF, which would be thefirst in nearly 2 decades, contrasts with a 3%increase requested by President George W.Bush. It also makes a mockery of a 15% an-nual rise called for by a 2001 law that, un-fortunately for scientists, appropriators don’thave to follow. “We’re still hopeful that thenumbers will improve after the Senate hasacted,” says acting NSF Director Arden Bement. “We are dealing with some frus-trated appropriators.”

The frustration stems from the fact thatNSF and NASA are part of a larger spend-ing bill that also funds the Veterans Admin-istration. Historically, veterans’ needs takeprecedence, especially in an election year.This year, the House panel approved an ex-

tra $2.5 billion for veterans’ health care,leaving little new money for other agencies.“We can’t compete with the veterans,” saysSam Rankin, chair of the Coalition for Na-tional Science Funding and a lobbyist for theAmerican Mathematical Society.

The House bill would trim NSF’s bread-and-butter research programs by $109 mil-lion, to $4.1 billion. It would cut the $935million education directorate by 10%, in-

cluding no new funding for a program thatlinks universities with local schools. Itwould also delay the start of the NationalEcological Observatory Network while al-lowing design work on two prototype sites.

For NASA, the $228 million cut reversesa Bush Administration proposal for a $1.1 billion increase for moon and Mars ex-ploration. Several new space science mis-sions took it on the chin while the committeepiled on millions of dollars in earmarks. Thepanel rejected the entire $70 million request-ed to begin a robotic lunar exploration effortand nixed $12.4 million to start the scientificwork on a Jupiter Icy Moons Orbiter andnearly all of the $17.6 million proposed foran Orbiting Carbon Observatory. At the sametime, the panel added goodies such as$150,000 for the Coca Cola Space ScienceCenter in Columbus, Georgia, and $3 millionfor the National Center of Excellence inBioinformatics in Buffalo, New York.

The committee also refused to fund a newCrew Exploration Vehicle that ultimately

could send humans to the moon and Mars.But it fully funded the $4.3 billion request forthe space shuttle. The panel said it backed theidea of Bush’s exploration vision but notedthat the committee “does not have sufficientresources.” The White House says it may vetothe bill if the NASA numbers don’t improve.

One agency that took a big hit last yearmay get a chance to climb partway out of itsbudget hole. On 8 July, the House approved

an 11% increase for the in-tramural programs at NISTand told the agency to spendwhatever it takes from its re-search account to outfit itsnew Advanced MaterialsLaboratory. Eighty-twoNIST employees have ac-cepted buyouts, and a better2005 budget, says Bement,who also heads NIST,means that “we won’t haveto lay off any scientists.”

At DOE, science advo-cates are praising a 25 JuneHouse vote giving theagency’s science off ice a3% boost to $3.6 billion, re-jecting a White House callfor a modest cut. Super-

computing research was a big winner, get-ting a 16% jump to $234 million. DOE’sheavily earmarked biological research ac-count, however, would slump 11% to $572million. Among the victims is a new $5 mil-lion molecular tag production facility. Law-makers said they didn’t like the depart-ment’s plan to allow only DOE labs to bidfor the project.

Defense researchers are pleased with an8% increase, to $1.5 billion, for basic re-search included in a Department of De-fense (DOD) spending bill to be signedsoon by Bush. That reverses a proposed5% cut. Applied research would jump 12%to nearly $5 billion.

Now, science advocates are waiting tosee how the Senate deals with other sciencebudgets. But many observers predict thatthe final numbers won’t be settled until latethis year, in a lame-duck session after theelections. –JEFFREY MERVIS

With reporting by David Malakoff and AndrewLawler.

Caught in a Squeeze Between Tax Cuts and Military SpendingU . S . S C I E N C E B U D G E T

–4%

–2%

0

2%

4%

6%

8%

10%

12%

% C

hang

e vs

. 200

4

2005 Science Budgets So Far

NIST

DOD Basic

Research

DOENIH

NASANSF

–2.5%

–1.5%

2.6%3.0%

8.0%

11.0%

Bleak house. Congress has approved only the DOD budget; funding

bills for other agencies are awaiting full House and Senate action.


CR

ED

IT:G

.LEG

NA

ME E

T A

L.

589

GAO Faults Science Agencieson Title IX ComplianceThree top U.S. science agencies havefailed to enforce a federal law aimed atincreasing female participation in educa-tional programs, according to a reportunveiled last week by the GovernmentAccountability Office (GAO), Congress’sinvestigative arm. The report, which wasrequested by Senators Ron Wyden(D–OR) and Barbara Boxer (D–CA) after a2002 hearing (Science, 11 October 2002,p. 356), says the Department of Energy,NASA, and the National Science Founda-tion have not been monitoring granteeinstitutions to check if they are comply-ing with Title IX. The 32-year-old legisla-tion, which allows the government towithhold funds from institutions thatpractice gender discrimination, applies toall fields of education, but its impact hasmostly been limited to athletics.

The GAO report confirms that thefederal government needs to enforce Title IX “not just on the playing field butalso in the classroom,” says Wyden. Hebelieves compliance reviews by grantingagencies are essential to close the gendergap in the sciences and engineering.Massachusetts Institute of Technology bi-ologist Nancy Hopkins, who chaired astudy on the status of women facultymembers at MIT’s School of Science, pre-dicts that “without government oversightand support, the full participation ofwomen and minorities in science and en-gineering will not occur in our lifetime—or in the lifetime of our children.”

–YUDHIJIT BHATTACHARJEE

Mexico Approves GenomicMedicine InstituteAfter 5 years of discussion, Mexico is get-ting a new institute for genomic medi-cine. President Vicente Fox last week ap-proved construction of the $200 millionINMEGEN center in Mexico City, which isexpected to employ 120 researchers andopen it first units next year.

The institute, which will focus in parton disease susceptibilities among Mexi-co's dozens of indigenous groups, will beled by biomedical researcher GerardoJiménez-Sánchez of Johns Hopkins Uni-versity in Baltimore, Maryland (Science,11 April 2003, p. 295).

“We cannot afford the luxury of notjoining this knowledge revolution,” saidFox. Jiménez-Sánchez says INMEGEN re-searchers will not work with human em-bryos. Mexican law allows both embryoresearch and therapeutic cloning.

–XAVIER BOSCH

ScienceScope

A bold set of prion experiments in micemay have proven that the misshapen pro-teins are, by themselves, infectious. If thework holds up, it will be a watershed inprion biology, validating the belief thatthese proteins alone are the culprits in“mad cow disease” and similar illnesses.

As is typical for the controversy-ladenfield, however, many scientists express reser-vations about the study on page 673. It wasled by Stanley Prusiner of the University ofCalifornia, San Francisco, who won the Nobel Prize in 1997 for discovering prions.

“It’s really a striking result that seems to

fill in one more piece of the infectivity puz-zle,” says Byron Caughey, a biochemist atthe National Institutes of Health’s RockyMountain Laboratories in Hamilton, Mon-tana. “But,” he adds, “it’s worth pointing outsome significant caveats.”

For years, biologists have tried to provethat a protein called PrP can misfold andbecome an infectious prion by purifyingprotein clumps from diseased brains andinjecting them into healthy animals. But ithasn’t been clear that PrP alone was whatwas being injected; using synthetic mis-folded PrP, meanwhile, hasn’t reliably trig-gered disease.

In their tests, Prusiner and colleagues usedtransgenic mice making 16 times the normalamount of PrP. These mice express a truncat-ed PrP that may more readily make up prionclumps. This, the group reasoned, might sen-sitize the animals to introduced PrP.

To obtain PrP free of brain tissue, Prusin-

er’s team genetically altered Escherichia colibacteria into producing PrP fragments thatthey misfolded to form amyloid fibrils, whichhave been implicated in various brain dis-eases. Prusiner’s team injected those prionfibrils into the brains of the mice.

Almost a year later, with no animalssick, the researchers were ready to declarethe study a failure. But then, 380 days afterbeing inoculated, one of the mice showedsymptoms of a prionlike disease. Eventual-ly, all seven inoculated mice showed neu-rological disease, the last one 660 days af-ter injection.

Prusiner’s team also inoculated a batch ofnormal animals with brain tissue from one ofthe sick ones. These rodents took about 150days to sicken.

“It is a spectacular breakthrough,” saysNeil Cashman, a neuroscientist at the Uni-versity of Toronto. “This is the beginningof the end of all the objections about theprion hypothesis.”

Not so fast, say some experts. Do Prusiner’s mice with excess PrP get sick nor-mally? wonders John Collinge, director ofthe Medical Research Council Prion Unit atUniversity College London. His team hadrelied on rodents with 10 times the normallevel of PrP but abandoned them after find-ing prion disease–like pathology in animalsthat hadn’t been inoculated with anything.

Prusiner’s mice, says Collinge, may be“poised” to become infectious even withoutthe inoculation; giving them a shot of syn-thetic, misfolded PrP may push them overthe edge, but so might other stresses.

The long latency time between inocula-tion and disease also worries prion experts.Some wonder if the experiments were con-taminated by other prion strains in the lab.Yale University neuropathologist LauraManuelidis, who has long criticized the pri-on hypothesis, says the brain samples fromsome of Prusiner’s mice resemble RMLscrapie, a common strain. Prusiner countersthat with contamination, the control animalsinoculated with saline should have gottensick as well.

Another explanation for long latency isthat infecting animals with synthetic PrP isinefficient. The first inoculations may havecontained few prions, says Prusiner. Thismight also explain why no one has yet ac-complished the gold-standard experiment:infecting normal mice, not transgenic ones,with pure prion proteins. Until then, one ofbiomedicine’s longest-running controversiesis likely to continue. –JENNIFER COUZIN

An End to the Prion Debate? Don’t Count on It

B I O M E D I C I N E

Building a prion? In a model of prion forma-

tion, misshapen PrP proteins (red) stack up into

amyloid fibrils.


MILWAUKEE, WISCONSIN—This week, archae-ologists will begin to dig 48 kilometerssouth of here, at a site that even skeptics saymay be the most convincing yet in demon-strating the early presence of humans in theAmericas. Scientists will search a muckylakeside just west of the city of Kenosha for additional remains of a woolly mam-moth. Bones found pre-viously bear marks fromhuman butchering andhave been dated to13,500 radiocarbon yearsbefore present—a full2000 years before big-game hunters known asthe Clovis people werethought to have arrivedon the continent.

Sites near Kenosha“may be the best pre-Clovis sites in NorthAmerica,” says teamleader Michael Waters ofTexas A&M Universityin College Station. Evenpre-Clovis skeptic Stuart Fiedel, an archae-ologist with the Louis Berger Group inWashington, D.C., agrees that “the Kenoshasites are high up on my radar screen. On theface of it, they seem to be one of the bestcases [of pre-Clovis evidence].”

Archaeologists long thought that Ameri-ca was first settled by the Clovis hunters,who crossed the Bering Strait and movedsouth through an ice-free corridor around11,500 radiocarbon years ago. Then in re-cent years dozens of sites in both North andSouth America pointed to an even older hu-man occupation. But each pre-Clovis site

has been bitterly contested (Science, 2March 2001, p. 1730), and a handful of in-fluential archaeologists believes that defini-tive pre-Clovis evidence is lacking. “One ofmy problems with the [pre-Clovis] positionis that the sites that it is founded on are still

dubious,” says Fiedel. Hence the excitement over the sites near

Kenosha. In 1990, an amateur archaeologistfound butcher marks on mammoth bonesstored at a local historical museum; archaeol-ogists later excavated at two sites, those of theSchaefer and Hebior mammoths. Thesemammoth bones are so well preserved thatcollagen could be extracted from inside thebone for radiocarbon dating, yielding dates ofabout 12,500 radiocarbon years ago, 1000years before the Clovis people. And a handfulof crude stone tools—unlike the elegantspear points of the Clovis people—were re-covered under the bone piles. All in all, thesites are unique, with “unequivocal stonetools [and] excellent dates,” says Waters.

Now his team is in pursuit of an evenolder Kenosha mammoth at Mud Lake,where a few bones with cut marks wereunearthed during a ditch-digging project inthe 1930s and later dated. Waters believesthat the rest of the mammoth is there andplans to try to relocate it this summerwhile scouting for new sites for future ex-cavations. The preliminary dig starts thisweek, but because heavy rains have slowedthe work, full-scale excavation of MudLake isn’t expected until next year.

Given the potential of the Kenosha sites,they have attracted little attention. “I reallydon’t understand why there has not beenmore investigation devoted to [them] todate,” says Fiedel. Starting this summer, Waters’s crew hopes to change that.

–TERRENCE FALK

Terrence Falk writes on science, education, andpublic policy from Milwaukee.

Wisconsin Dig Seeks to Confirm Pre-Clovis Americans

A R C H A E O L O G Y

N E W S O F T H E W E E K

States Sue Over Global WarmingIn a legal gambit aimed against global warm-ing, the attorneys general of eight states lastweek sued the five largest emitters of carbondioxide in the United States for creating apublic nuisance. The states are asking that theelectric utility companies cut emissions by3% each year for a decade. Legal experts pre-dict the states’ case will be an uphill battle.

Carbon dioxide litigation is heating up. In2002, environmental groups sued the Over-seas Private Investment Corp. and the Export-Import Bank of the United States for not con-ducting environmental reviews on the powerplants they financed. And last year, Maine,Massachusetts, and Connecticut sued the En-vironmental Protection Agency for not regu-lating CO2 as a pollutant under the Clean AirAct. Now, the states have taken the first legalaction directly against CO2 emitters.

The plaintiffs—California, Connecticut,Iowa, New Jersey, New York, Rhode Island,Vermont, and Wisconsin, along with the

City of New York—claim that the CO2 thatutility companies release contributes toglobal warming, which will harm state resi-dents. The alleged ills include increasednumbers of deaths from heat waves, moreasthma from smog, beach erosion, contami-nation of groundwater from rising sea level,and more droughts and floods. “The harm toour states is increasing daily,” Eliot Spitzer,the attorney general of New York state, saidat a press conference.

The defendants together spew about 650million tons of CO2 a year. Their 174 fossilfuel–burning plants contribute roughly 10%of the anthropogenic CO2 in the UnitedStates. The suit maintains that annual cuts of3% are feasible through making plants moreefficient, promoting conservation, and usingwind and solar power—without substantiallyraising electric bills. “All that is now lackingis action,” Spitzer said.

That claim irks American Electric Power

of Columbus, Ohio, a defendant. Spokes-person Melissa McHenry says that the com-pany had already committed to reducing itsemissions by 10% by 2006. “Filing lawsuitsis not constructive,” she says. “It’s a globalissue that can’t be addressed by a smallgroup of companies.”

It will also be a tough suit to win, saysRichard Brooks of Vermont Law School inSouth Royalton, who studies the legal issuesof air pollution. The fact that global warm-ing is a planetwide phenomenon will makeit diff icult to establish how much thesecompanies are contributing to the claimedharm. And under public-nuisance law, theplaintiffs must show that their citizens aresuffering significantly more than the nationas a whole. “I would be totally amazed ifthe court gave this a serious response,”Brooks says. “This makes me imagine thatthis is more of a symbolic suit.”

–ERIK STOKSTAD

E N V I R O N M E N T

CR

ED

IT:C

OU

RT

ESY

OF

TH

E K

EN

OSH

A P

UB

LIC

MU

SEU

M

Mammoth meal? Bones from Kenosha, dated to 12,500 radiocarbon

years ago, show signs of butchery by early Americans.


CR

ED

IT:J

AM

ES C

HA

TT

ER

S

ScienceScope

591

Panel Pans UC-Novartis DealThe University of California (UC), Berkeley,should pass on any proposals similar to theagreement it once made with pharmaceu-tical giant Novartis, according to a new in-dependent report commissioned by theschool’s academic senate. In 1998, Novartispledged $25 million over 5 years to theplant and microbial biology department inexchange for significant access to the de-partment’s labs and scientific discoveries.The agreement was greeted with outrageby many researchers at the school andacross the country (Science, 17 January2003, p. 330).

The direct impacts of the pact on theuniversity “have been minimal,” concludesthe report, authored by food and agricul-tural specialist Lawrence Busch of MichiganState University in East Lansing and col-leagues.Although graduate students in thefield enjoyed increased stipends, “few or nobenefits” in terms of patent rights or in-come went to the university or to Novartisand its successor Syngenta, according tothe report. Busch and his co-authors alsoconcluded that the agreement did notdamage the department’s basic science ef-forts, as many opponents feared it would.But the report recommends that Berkeleyavoid future industry agreements “that in-volve complete academic units or largegroups of researchers” and urges “broad de-bate early in the process of developing newresearch agendas.”The study will be sub-mitted to the Berkeley Senate on 1 Augustfor consideration. –ANDREW LAWLER

House Cuts EPA R&D,Restores STAR GrantsResearch budgets at the EnvironmentalProtection Agency (EPA) would face a4.3% cut, to $589 million, in a spendingplan approved last week by the House ap-propriations committee. The cuts are partof an increasingly gloomy budget picturefor science (see p. 587).

Environmental researchers did getsome good news, however.The panel re-stored funding to EPA’s extramural grantsprogram, called Science to Achieve Results(STAR), bringing the program back to itsfiscal year 2004 level of about $76.1 mil-lion, with an additional $9.5 million spenton graduate fellowships. In February, theBush Administration proposed deep cutsto both STAR grants and the fellowships.The House support for STAR is encourag-ing, says Craig Schiffries, director of sci-ence policy at the National Council for Sci-ence and the Environment in Washington,D.C., but he’s disappointed that the fund-ing is still lower than the $100 million re-quested in recent years. –ERIK STOKSTAD

When Native American tribes decided lastweek not to fight an appeals court ruling, itlooked as though the way was clear for sci-entists to study the 9300-year-old skeletoncalled Kennewick Man, which has been tiedup in legal battles for the past 8 years. Butscientists say that although the ruling sets afavorable precedent for studying other an-cient skeletons, they are not optimistic aboutgetting to study Kennewick Man himselfanytime soon. The government continues tofind fault with outside scientists’ researchplans and to deny access to the remains. Ne-gotiations are in progress, but the lawyer forthe eight scientist-plaintiffs in the suit, AlanSchneider of Portland, Oregon, says, “we arestill far apart.” Going back to court, he adds,“is definitely a possibility.”

The Kennewick case finally appeared tohave come to an end on 19 July when thedefendants, four tribal groups, decided notto appeal to the U.S. Supreme Court a deci-sion by the 9th U.S. Circuit Court of Ap-peals. That court ruled that because there isno evidence linking the Kennewick skeletonto any existing tribe, the Native AmericanGraves Protection and Repatriation Act(NAGPRA) does not apply to it (Science, 13 February, p. 943). Thecourt’s interpretation ofNAGPRA is a significantadvance that will have“major implications” inother cases in which NativeAmerican groups areclaiming remains, saysRobson Bonnichsen of TheCenter for the Study of theFirst Americans at TexasA&M University in Col-lege Station. In a U.S.Army Corps of Engineersproject in Texas, for exam-ple, he says, Native Ameri-cans at f irst claimed re-mains from a 4000-year-old burial ground, but acompromise has been reached so that scien-tists will have access to them.

Meanwhile, scientists are eager to studyKennewick Man, one of the oldest skeletonsin North America. Schneider says that in2002, the scientists submitted a 40-pagestudy plan to the Department of the Interiorand the Corps of Engineers, which has cus-tody of the remains at the Burke Museum inSeattle. It is “a state-of-the-art proposal todo the most detailed look at a first Americanthat has ever been put together,” says Bon-nichsen. “We wanted to do a class act.”

But officials at Interior and the Corps of

Engineers have responded with a throng ofobjections. According to Bonnichsen, theCorps of Engineers says the skeleton is “frag-ile,” and officials seek to limit the number ofscientists who have access to it. “The corps isconcerned about the condition and wants tolimit handling to what is needed to producenew knowledge,” says Frank McManamon,chief archaeologist at the National Park Ser-vice. McManamon, who has been advisingon the government response to the study plan,says the plan doesn’t “build on the substantialamount of scientific investigation that has al-ready been done” by government-appointedscientists. For example, he says that Bonnich-sen and colleagues want to take bone samplesfor DNA testing, even though sampling hasalready been done and three separate labscouldn’t extract any DNA.

Lawyer Schneider counters that the government-sponsored radiocarbon andDNA tests “used or damaged up to 60grams of the skeleton,” whereas the scien-tists have proposed “microsampling,” whichwould destroy no more than 1.5 grams ofbone. He adds that many other areas needstudy. For example, although government-appointed scientists did computed tomogra-

phy (CT) scans to examine the projectilepoint lodged in the skeleton’s pelvis, Schnei-der says that “there is still a major controver-sy over which direction [it] entered,” andthat more sophisticated CT technology isnow available to study it. “What Frank [McManamon] seems to be saying is ‘We’velooked at them, so you don’t need to’ ”—hardly a scientific stance, says Schneider.

While the haggling continues, NativeAmericans have indicated that they willnow embark on a nationwide campaign topressure Congress to rewrite NAGPRA.

–CONSTANCE HOLDEN

Court Battle Ends, Bones Still Off-LimitsK E N N E W I C K M A N

Still fighting. Scientists seek access to the bones of Kennewick

Man, who died with a projectile point in his pelvis (arrow).


CR

ED

IT:(

RIG

HT

) D

EN

NIS

CO

OK

/AP P

HO

TO

S

For the past several months, the Bush Ad-ministration has responded with strong de-nials to charges that it has chosen membersof scientific advisory committees in part fortheir political views. The charges are eitherwrong or distorted or they reflect aberrationsin the selection process, Ad-ministration officials haveasserted (Science, 16 July, p. 323). But last week aprominent House membertook a different tack: There’snothing wrong with mixingscience and politics in deter-mining the makeup of sci-entif ic advisory commit-tees, says RepresentativeVern Ehlers (R–MI).

Ehlers, a former physicsprofessor and staunch con-servative, offered thisview in an impromptu de-bate with RepresentativeHenry Waxman (D–CA), adyed-in-the-wool liberal,at a meeting of the Nation-al Academies’ Committeeon Science, Engineering, and Public Poli-cy. The committee is taking its third stabat recommending how the governmentcan improve its pool of scientif ic andtechnology talent. Its previous reports fo-cused on ways to make full-time jobs inWashington, D.C., more welcoming toscientists, but this year’s effort is also ex-amining the hundreds of outside advisorycommittees that help federal agencies dotheir work. The panel, which includes vet-erans from previous administrations span-ning both parties, hopes to deliver its re-port soon after the November election.

The problems flagged in earlier reportsstill exist: an intrusive and time-consumingvetting process, a likely cut in salary, andthe uncertainty of winning Senate confir-mation. Panel chair John Edward Porter, aformer representative from Illinois andpatron of the National Institutes ofHealth, says the issues remain “intract-able.” But Porter’s f irst question to hisformer colleagues signaled that, this timearound, the burning questions are morepolitical than logistical. “Do you thinkthat it’s appropriate for the government toask someone being considered for an ad-visory position, ‘Who did you vote for?’ ”Porter wanted to know.

“I think that it’s an appropriate questionto ask,” replied Ehlers, who also defended

the practice of asking where potential advis-ers stand on various hot-button issues.“Abortion is not a scientific issue, and yetthere are technical committees that give ad-vice on issues relating to abortion, like theuse of embryonic stem cells in research,” he

said. “The dividing line [between politicsand science] is not clear. My first principleis to make sure that all views are representedat the table, to get the best people, and thenlet them shout at each other. That’s the idealscientific advisory committee.”

Waxman rejected that argument. “Thereis a line you need to draw,” he insisted.“For political appointees, the presidentshould expect that his nominee supportshis policies. But for advisory committees,they ought not to ask one’s views on abor-

tion, or how they voted [in the 2000presidential election].” Waxman laterinsisted that the Bush Administrationhas imposed its own judgments onthe advisory process, “settling on apolicy first and then finding scien-tists to support that view.”

Earlier in the day, presidential sci-ence adviser John Marburger told thepanel that the candidate “pool isalarmingly small” when it comes tohiring good federal science man-agers. But he dodged a question fromone of his forerunners, Frank Press,about interference from the WhiteHouse in staffing his Office of Sci-ence and Technology Policy. Resist-ing such intrusions, Marburger said,“is easier than you might think.”

–JEFFREY MERVIS

Congressmen Clash on Politics andScientific Advisory Committees

U . S . S C I E N C E P O L I C Y

N E W S O F T H E W E E K

Los Alamos Suspends 19 EmployeesThe Department of Energy’s (DOE’s) secu-rity and safety problems continue to esca-late. George “Pete” Nanos, head of LosAlamos National Laboratory in New Mex-ico, last week suspended 19 employees—including some senior scientists—pendingan investigation of possible rules viola-tions. He had already shut down virtuallyall work at the lab until the investigation iscompleted (Science, 23 July, p. 462). Then,starting this week, DOE Secretary SpencerAbraham suspended classified work in-volving portable computer disks at allDOE facilities, including Lawrence Liver-more National Laboratory in California.

The massive “stand down” is needed,Abraham says, to make sure that securitylapses at Los Alamos weren’t repeated else-where and to “make certain that specific in-dividuals can be held responsible and ac-countable for future problems.”

Both moves are rooted in a 7 July in-ventory at Los Alamos that concluded thattwo classified disks were improperly re-moved from a safe. Then, on 14 July, an in-tern’s eye was injured by a research laserthat had not been turned off. Furious aboutthe incidents, Nanos suspended research at

the laboratory and warned that he wouldfire “cowboys” who flouted the rules.

On 22 July, citing “almost suicidal de-nial” of security and safety practices,Nanos suspended 15 employees involvedin the loss of the disks, along with four in-volved in the laser accident. All will con-tinue to receive pay but are barred fromentering the laboratory without an escort.The FBI is investigating the lost disks, andNanos said some employees could facecriminal charges.

The DOE-wide shutdown is affectingabout a dozen laboratories that do classi-fied work. None of the labs will be able toresume activity until they have performeda series of steps, including a complete in-ventory of portable disks, the creation ofsecure repositories for disks and other re-movable devices containing classified in-formation, and a visit from an independentreview team.

In the meantime, some researchers arebecoming frustrated. In Los Alamos, for in-stance, residents report a growing number ofcars sporting ironic bumper stickers that say“Striving for a Work-Free Safe Zone.”

–DAVID MALAKOFF

N AT I O N A L L A B S

En garde. Vern Ehlers (left) says it is appropriate to ask potential panel

members whom they voted for; Henry Waxman disagrees.


CR

ED

ITS:(

TO

P)

C.S

.GO

LDSM

ITH

AN

D J

.M.K

AT

Z/C

DC

;EY

E O

F SC

IEN

CE/P

HO

TO

RESEA

RC

HER

S I

NC

.

Researchers say crossing avian and human flu viruses is crucial to understanding the threat of a new influenzapandemic, but they admit that they might create a monster

Tiptoeing Around Pandora’s Box

News Focus

Once again, the world is crossing its fingers.The avian influenza outbreak in Asia, al-ready one of the worst animal-health disas-ters in history, has flared up in four coun-tries; tens of thousands of birds are beingkilled in desperate attempts to halt thevirus’s spread. And again, the unnervingquestion arises: Could the outbreak of theH5N1 strain spiral into a human flu pan-demic, a global cataclysm that couldkill millions in a matter of monthsand shake societies to their core?

There is a way to find out, flu sci-entists say—but it’s controversial.

Leaving nature to take its course, apandemic could be ignited if avian andhuman influenza strains recombine—say, in the lungs of an Asian farmerinfected with both—producing abrand-new hybrid no human is im-mune to. By mixing H5N1 and hu-man flu viruses in the lab, scientistscan find out how likely this is, andhow dangerous a hybrid it would be.

Such experiments can give theworld a better handle on the risks, butthey could also create dangerous newviruses that would have to be de-stroyed or locked up forever in a sci-entific high-security prison. An acci-dental release—not so far-fetched ascenario given that the severe acuterespiratory syndrome (SARS) virusmanaged to escape from three Asianlabs in the past year—could lead toglobal disaster. Given their scientificmerit, the World Health Organization(WHO) is enthusiastically promoting the ex-periments. But worried critics point out thatthere is no global mechanism to ensure thatthey are done safely.

Despite the concerns, such studies havealready begun. In 2000, the U.S. Centers forDisease Control and Prevention (CDC) inAtlanta, Georgia, started experiments to cre-ate crossovers between the H5N1 strain iso-lated during a 1997 outbreak in Hong Kongand a human flu virus adapted for the lab.The study was suspended when CDC’s fluresearchers became overwhelmed by SARSand the new H5N1 outbreak, both in 2003,says CDC flu expert Nancy Cox, who ledthe work. But the agency plans to resume

the work shortly with the H5N1 strain nowraging in Asia.

Others are exploring the options as well.Virologist Albert Osterhaus of Erasmus Uni-versity in Rotterdam, the Netherlands, is ea-ger to try not just H5N1 but also other birdflu strains, such as H7N7. The Netherlandswon’t have the required high-level biosafetylab until late 2005, so Osterhaus is talking to

researchers in France who do. In the UnitedKingdom, researchers at the Health Protec-tion Agency, the National Institute for Bio-logical Standards and Control, and universi-ties are also discussing the idea. There areno concrete plans yet—in part because of alack of funds—but there’s a consensus thatthe studies are important and that Britain iswell suited to do them, says influenza re-searcher Maria Zambon of the Health Pro-tection Agency.

The aim of reassortment studies, as they’recalled, would not be to develop new counter-measures, says WHO’s principal flu scientist,Klaus Stöhr, because researchers believe cur-rent drugs and an H5N1 vaccine in develop-ment would work against a pandemic strain as

well. But the experiments would provide abadly needed way to assess the risk of a pan-demic. If they indicate that a pandemic virusis just around the corner, health officialswould further intensify their fight in Asia andgo full-throttle in stashing vaccines and drugs;if not, they could breathe a little easier. “It’s anextremely important question, and we have aresponsibility to answer it,” insists Stöhr.

The safety worries are legitimate,Stöhr concedes, and the work shouldbe done only by labs with ample flu expertise and excellent safety systems—not the ones that let SARSout. “We don’t want people just fid-dling around,” he says. He also down-plays concerns that the results, whenpublished, might help those who wouldunleash a pandemic on purpose. Any-one with the scientific smarts to do socan already find plenty of ideas in theliterature, Stöhr asserts. Moreover, thestudies are unlikely to produce any-thing that could not arise naturally,says Osterhaus: “You could create amonster. But it’s a monster that naturecould produce as well.”

But critics beg to differ. “We’vebeen debating whether to destroy thesmallpox virus for years—and nowwe’re planning to create somethingthat’s almost as dangerous?” asks MarkWheelis, an arms-control researcher atthe University of California, Davis.Wheelis also points out that there’s noway to keep countries with poor safety

records from getting in on the game. At thevery least, there should be some global con-sensus on how to proceed, adds Elisa Harris,a researcher at the Center for Internationaland Security Studies at the University ofMaryland, College Park—although no formalmechanism for reaching it exists.

Mix and match

The H5N1 strain has been vicious to its hu-man victims, killing 23 of 34 patients inVietnam and Thailand this year. So far, how-ever, every known patient had been in con-tact with infected birds; there’s no evidencethat the virus can jump from one person tothe next—for now. But the virus couldevolve inside one of its human hosts, acquir-

Risk assessment. The H5N1 influenza strain is highly lethal to

humans, but whether it could trigger a pandemic is still uncertain.


CR

ED

ITS:(

TO

P T

O B

OT

TO

M)

C.S

.GO

LDSM

ITH

AN

D J

.M.K

AT

Z/C

DC

;P.H

UEY

/SC

IEN

CE

ing mutations that make it possibleto infect humans directly, Stöhrsays. Another scenario—one re-searchers believe sparked severalprevious influenza pandemics—isreassortment with a human flu virusin a person infected with both.

Influenza has a peculiar genomethat’s divided into eight loose seg-ments, most of them containing pre-cisely one gene. Each segment iscopied separately in the host cell’snucleus; at the end of the reproduc-tion cycle, all eight meet up withone another—and with envelope andmembrane proteins—to form a newvirus particle that buds from the hostcell membrane to wreak havoc else-where. When a cell happens to beinfected with two different strains,homologous segments can mix andmatch into new, chimeric viruses.

To create a worldwide outbreak, anewcomer must cause disease in humans andbe transmissible between them, and its coatmust look so new that no human immunesystem recognizes it. This is determined pri-marily by the two glycoproteins on the viralsurface, hemagglutinin and neuraminidase—the “H” and “N” in names like H5N1.(Hemagglutinin comes in at least 16 differenttypes, N in nine.) The current fear is that theAsian flu will keep its H5—which humanshave never seen before—but swap enough ofthe remaining seven gene segments withthose of a human strain to become moreadept at replication in its new host.

During H5N1’s first major outbreak inHong Kong poultry in 1997, 18 people gotsick and six died. But the outbreak wasstamped out efficiently, and little was heard ofH5N1 for 6 years—until it came roaring backlast year. Given the magnitude of the currentoutbreak, the riddle is why reassortment hasnot yet taken place, says Stöhr. Reassortmentstudies could help explain whether the worldhas simply been lucky, or whether there’ssome barrier to reassortment of H5N1.

The experiments are straightforward. Re-searchers take a cell line such as MDCK orVero cells, often used for virus isolation, andadd both H5N1 and a currently circulatinghuman strain, such as H3N2 or H1N1. Orthey can use a slightly less natural techniquecalled reverse genetics, with which virtuallyany combination of genes can be put into aflu virus. Any viable hybrid strains would beinoculated into mice; those that cause dis-ease would move on to ferrets, a speciesvery similar to humans in its susceptibilityto influenza. Any strain that is pathogenic in

ferrets and also jumps, say, from a sick ani-mal to a healthy one in an adjacent cagecould be humankind’s next nightmare.

During its first round of experiments withthe H5N1 strain, CDC managed to createseveral reassortants, Cox says, but it didn’tget around to characterizing them; they’restill sitting in a locked freezer in Atlanta.

Global risks, global review?

Most agree that such experiments are in aleague of their own. Controversial flu studieswere conducted in the past; for instance, re-searchers sequenced parts of the genome ofthe “Spanish flu” strain from 1918 (Science,21 March 1997, p. 1793) and inserted itsgenes into other strains to find out why itwas so deadly. But that didn’t amount to awholesale fishing expedition for pandemicstrains. And because the 1918 strain was anH1 virus, just like one of the currently activeones, you’d expect at least some immunityto it in the human population, says YoshihiroKawaoka of the University of Tokyo and theUniversity of Wisconsin, Madison, whostudies the 1918 strain. With an H5 virus, incontrast, everyone would be vulnerable.

Yet although most countries have sys-tems to review the safety and ethical aspectsof run-of-the-mill scientific studies, nonehave formal panels to weigh studies thatcould, say, put the entire world at risk or beof potential help to bioterrorists. [The U.S.government has announced plans for a na-tional biosecurity panel and a review systemto fill that gap (Science, 12 March, p. 1595),but they have yet to be implemented.] So al-though CDC’s first round of studies cleared

all the usual review hurdles at theagency, Cox says, nothing beyond thatwas considered necessary.

Since then, “the times havechanged,” Cox says. The H5N1 strainnow plaguing Asia, with which CDCwants to work this time, appears to bemore virulent than the 1997 version, andthe specter of nefarious use ofpathogens looms much larger. More-over, the mishaps with SARS have madepeople jittery about labs’ abilities to

keep bugs on the inside. That’s whyCox says she has consulted moreextensively with colleagues insideand outside CDC, including ex-perts such as Nobel laureateJoshua Lederberg and WHO. She

also plans to seek approval fromcolleagues at the U.S. National Insti-tutes of Health and the U.S. Foodand Drug Administration.

But flu researcher Karl Nichol-son of the University of Leicester, U.K., saysthere should be a more formal, global con-sensus on the necessity of the studies, whoshould conduct them, and how. For anycountry to undertake them on its own, hesays, “is like a decision to start testing nu-clear weapons unilaterally.” WHO would bethe best organization to start such a process,says Harris: The destruction of the smallpoxvirus has been debated at WHO, and an in-ternational panel there is overseeing experi-ments with it at CDC and in Russia.

But Stöhr believes existing safeguardssuffice. The studies have been discussedwidely with scientists in WHO’s global flulab network and at a recent flu meeting inLisbon, he says, and have met with nothingbut “overwhelming agreement.” “If there areother voices, we will take them seriously,”Stöhr adds—but for now, it’s up to the labsto have their plans rigorously vetted by na-tional authorities and get started.

Eventually, any strain with pandemic po-tential should be destroyed, he says. Butthere’s no way to enforce this, and skepticspoint out that the smallpox virus was slatedfor destruction, too—until the threat ofbioterrorism created a movement to keep italive, perhaps indefinitely, for defensivestudies. In a way this discussion is moot,says Richard Webby of St. Jude Children’sResearch Hospital in Memphis, Tennessee.With flu strains readily available, anyonewith a good knowledge of molecular biolo-gy could recreate a pandemic virus once it’sdiscovered, he says. “You can destroy thisvirus,” Webby says, “but it will never reallybe gone.” –MARTIN ENSERINK

Two can tango. Flu virus genomes consist of eight segments,

each of which is copied separately by the host cell (left). When

two strains infect one cell, they can reassort (right).


CR

ED

IT:A

SSO

CIA

TED

PR

ESS

When he was 17 years old, Christopher Sim-mons persuaded a younger friend to helphim rob a woman, tie her up with electricalcable and duct tape, and throw her over abridge. He was convicted of murder and sen-tenced to death by a Missouri court in 1994.In a whipsaw of legal proceedings, the Mis-souri Supreme Courtset the sentenceaside last year. Now27, Simmons couldagain face execution:The state of Missourihas appealed to havethe death penalty re-instated. The U.S.Supreme Court willhear the case in Oc-tober, and its deci-sion could well reston neurobiology.

At issue iswhether 16- and 17-year-olds who com-mit capital offensescan be executed orwhether this wouldbe cruel and unusualpunishment, bannedby the Constitution’seighth amendment.In a joint brief filedon 19 July, eightmedical and mentalhealth organizations including the AmericanMedical Association cite a sheaf of develop-mental biology and behavioral literature tosupport their argument that adolescentbrains have not reached their full adult po-tential. “Capacities relevant to criminalresponsibility are still developing whenyou’re 16 or 17 years old,” says psychologistLaurence Steinberg of the American Psy-chological Association, which joined thebrief supporting Simmons. Adds physicianDavid Fassler, spokesperson for the Ameri-can Psychiatric Association (APA) and theAmerican Academy of Child and Adoles-cent Psychiatry, the argument “does not ex-cuse violent criminal behavior, but it’s animportant factor for courts to consider”when wielding a punishment “as extremeand irreversible as death.”

The Supreme Court has addressed some

of these issues before. In 1988, it held that itwas unconstitutional to execute convicts un-der 16, but it ruled in 1989 that states werewithin their rights to put 16- and 17-year-oldcriminals to death. Thirteen years later, it de-cided that mentally retarded people shouldn’tbe executed because they have a reduced ca-

pacity for “reasoning,judgment, and controlof their impulses,”even though they gen-erally know rightfrom wrong (see side-bar on p. 599). That isthe standard Sim-mons’s lawyers nowwant the court to ex-tend to everyone un-der 18.

Cruel and unusual?

Simmons’s lawyersargue that adolescentsare not as morallyculpable as adults andtherefore should notbe subject to thedeath penalty. Theyclaim that this viewreflects worldwide“changing standardsof decency,” a trendthat has been recog-nized in many U.S.

courts. Today, 31 states and the federal gov-ernment have banned the juvenile deathpenalty. The latest to do so, Wyoming andSouth Dakota, considered brain develop-ment research in their decisions.

Putting a 17-year-old to death for capi-tal crimes is cruel and unusual punish-ment, according to this reasoning. “Whatwas cruel and unusual when the Constitu-tion was written is different from today. Wedon’t put people in stockades now,” saysStephen Harper, a lawyer with the JuvenileJustice Center of the American Bar Associ-ation (ABA), which also signed an amicuscuriae brief. “These standards mark theprogress of a civilized society.”

The defense is focusing on the “culpa-bility of juveniles and whether their brainsare as capable of impulse control, decision-making, and reasoning as adult

brains are,” says law professor StevenDrizin of Northwestern University inChicago. And some brain researchers an-swer with a resounding “no.” The brain’sfrontal lobe, which exercises restraint overimpulsive behavior, “doesn’t begin to ma-ture until 17 years of age,” says neurosci-entist Ruben Gur of the University ofPennsylvania in Philadelphia. “The verypart of the brain that is judged by the legalsystem process comes on board late.”

But other researchers hesitate to applyscientists’ opinions to settle moral and legalquestions. Although brain research shouldprobably take a part in policy debate, it’sdamaging to use science to support essen-tially moral stances, says neuroscientist PaulThompson of the University of California,Los Angeles (UCLA).

Shades of gray

Structurally, the brain is still growing and ma-turing during adolescence, beginning its finalpush around 16 or 17, many brain-imagingresearchers agree. Some say that growth max-es out at age 20. Others, such as Jay Giedd ofthe National Institute of Mental Health(NIMH) in Bethesda, Maryland, consider 25the age at which brain maturation peaks.

Various types of brain scans andanatomic dissections show that as teensage, disordered-looking neuron cell bodiesknown as gray matter recede, and neuronprojections covered in a protective fattysheath, called white matter, take over. In1999, Giedd and colleagues showed thatjust before puberty, children have a growthspurt of gray matter. This is followed bymassive “pruning” in which about 1% ofgray matter is pared down each year duringthe teen years, while the total volume ofwhite matter ramps up. This process isthought to shape the brain’s neural connec-tions for adulthood, based on experience.

In arguing for leniency, Simmons’s sup-porters cite some of the latest research thatpoints to the immaturity of youthful brains,such as a May study of children and teens, ledby NIMH’s Nitin Gogtay. The team followed13 individuals between the ages of 4 and 21,performing magnetic resonance imaging(MRI) every 2 years to track changes in thephysical structure of brain tissue. As previousresearch had suggested, the frontal lobes ma-tured last. Starting from the back of the head,“we see a wave of brain change moving for-ward into the front of the brain like a forestfire,” says UCLA’s Thompson, a co-author.The brain changes continued up to age 21,the oldest person they examined. “It’s quitepossible that the brain maturation peaks afterage 21,” he adds.

The images showed a rapid conversion

Crime, Culpability, and the Adolescent BrainThis fall, the U.S. Supreme Court will consider whether capital crimes by teenagers under18 should get the death sentence; the case for leniency is based in part on brain studies

Neurosc ience

Test case. Christopher Simmons received the

death penalty for a crime he committed at 17.

from gray to white matter. Thompson saysthat researchers debate whether teens areactually losing tissue when the gray matterdisappears, trimming connections, or justcoating gray matter with insulation. Imag-ing doesn’t provide high enough resolutionto distinguish among the possibilities, henotes: “Right now we can image chunks ofmillions of neurons, but we can’t look atindividual cells.” A type of spectroscopythat picks out N-acetylaspartate, a chemi-cal found only in neurons, shows promisein helping to settle the issue.

In addition to growing volume, brainstudies document an increase in the organi-zation of white matter during adolescence.The joint brief cites a 1999 study by TomásPaus of McGill University in Montreal andcolleagues that used structural MRI to showthat neuronal tracts connecting different re-gions of the brain thickened as they werecoated with a protective sheath of myelinduring adolescence (Science, 19 March1999, p. 1908).

In 2002, another study revealed thatthese tracts gained in directionality as well.Relying on diffusion tensor MRI, which fol-lows the direction that water travels, Vin-cent Schmithorst of the Children’s HospitalMedical Center in Cincinnati, Ohio, andcolleagues watched the brain organize itselfin 33 children and teens from age 5 to 18.During adolescence, the tracts funneled upfrom the spinal tract, through the brainstem,and into motor regions. Another linked thetwo major language areas. “The brain isgetting more organized and dense withage,” Schmithorst says.

Don’t look at the light

Adults behave differently not just becausethey have different brain structures, ac-cording to Gur and others, but becausethey use the structures in a different way. Afully developed frontal lobe curbs impuls-es coming from other parts of the brain,Gur explains: “If you’ve been insulted,your emotional brain says, ‘Kill,’ but yourfrontal lobe says you’re in the middle of acocktail party, ‘so let’s respond with a cut-ting remark.’ ”

As it matures, the adolescent brain slowlyreorganizes how it integrates informationcoming from the nether regions. Using func-tional MRI—which lights up sites in thebrain that are active—combined with simpletests, neuroscientist Beatriz Luna of the Uni-versity of Pittsburgh has found that the brainswitches from relying heavily on local re-gions in childhood to more distributive andcollaborative interactions among distant re-gions in adulthood.

One of the methods Luna uses to probebrain activity is the “antisaccade” test: asimplified model of real-life responses de-

signed to determine how well the prefrontalcortex governs the more primitive parts ofthe brain. Subjects focus on a cross on ascreen and are told that the cross will dis-appear and a light will show up. They aretold not to look at the light, which is diffi-cult because “the whole brainstem is wiredto look at lights,” says Luna.

Adolescents can prevent themselvesfrom peeking at the light, but in doing sothey rely on brain regions different fromthose adults use. In 2001, Luna and col-leagues showed that adolescents’ prefrontalcortices were considerably more active thanadults’ in this test. Adults also used areas in

the cerebellum important for timing andlearning and brain regions that prepare forthe task at hand.

These results support other evidenceshowing that teens’ impulse control is noton a par with adults’. In work in press inChild Development, Luna found that vol-unteers aged 14 years and older performjust as well on the task as adults, but theyrely mainly on the frontal lobe’s prefrontalcortex, whereas adults exhibit a more com-plex response. “The adolescent is usingslightly different brain mechanisms toachieve the goal,” says Luna. Although thework is not cited in the brief, Luna says it


CR

ED

IT:P

AU

L T

HO

MPSO

N,K

IRA

LEE H

AYA

SH

I,A

ND

AR

TH

UR

TO

GA

/UC

LA,N

.GO

GTA

Y A

ND

JU

DIT

H R

APO

PO

RT

/NIM

H

Maturing brain. An NIMH study of 13 indi-

viduals over a decade reveals a process—

still under way in the late teens—in which

gray matter is replaced throughout the cor-

tex, starting at the rear.

Normal Brain Development

N E W S F O C U S

clearly shows that “adolescents cannot beviewed at the same level as adults.”

Processing fear

Other studies—based on the amygdala, abrain region that processes emotions, andresearch on risk awareness—indicate thatteenagers are more prone to erratic behav-ior than adults. Abigail Baird and DeborahYurgelun-Todd of Harvard Medical Schoolin Boston and others asked teens in a 1999study to identify the emotion they perceivein pictures of faces. As expected, functionalMRI showed that in both adolescents andadults, the amygdala burst with activitywhen presented with a face showing fear.But the prefrontal cortex didn’t blaze inteens as it did in adults, suggesting thatemotional responses have little inhibition.In addition, the teens kept mistaking fearfulexpressions for anger or other emotions.

Baird, now at Dartmouth College inHanover, New Hampshire, says that subse-

quent experiments showed that in teenagersthe prefrontal cortex buzzes when they viewexpressions of people they know. Also, thechildren identified the correct emotion morethan 95% of the time, an improvement of20% over the previous work.

The key difference between the results,says Baird, is that adolescents pay attention tothings that matter to them but have difficultyinterpreting images that are unfamiliar orseem remote in time. Teens shown a disco-erapicture in previous studies would say, “Oh,he’s freaked out because he’s stuck in the’70s,” she says. Teens are painfully aware ofemotions, she notes.

But teens are really bad at the kind ofthinking that requires looking into the futureto see the results of actions, a characteristicthat feeds increased risk-taking. Baird sug-gests: Ask someone, “How would you like toget roller skates and skate down some reallybig steps?” Adults know what might happenat the bottom and would be wary. But teens

don’t see things the same way, because “theyhave trouble generating hypotheses of whatmight happen,” says Baird, partly becausethey don’t have access to the many experi-ences that adults do. The ability to do soemerges between 15 and 18 years of age, shetheorizes in an upcoming issue of the Pro-ceedings of the Royal Society of London.

Luna points out that the tumultuous na-ture of adolescent brains is normal: “Thistransition in adolescence is not a disease oran impairment. It’s an extremely adaptiveway to make an adult.” She speculates thatrisk-taking and lowered inhibitions provide“experiences to prune their brains.”

With all the pruning, myelination, and re-organization, an adolescent’s brain is unsta-ble, but performing well on tests can maketeens look more mature than they are. “Yes,adolescents can look like adults. But putstressors into a system that’s already fragile,and it can easily revert to a less maturestate,” Luna says.

The amicus curiae brief endorsed bythe APA and others also describes thefragility of adolescence—how teens aresensitive to peer pressure and can be com-promised by a less-than-pristine childhoodenvironment. Abuse can affect how nor-mally brains develop. “Not surprisingly,every [juvenile offender on death row] hasbeen abused or neglected as a kid,” saysABA attorney Harper.

Biology and behavior

Although many researchers agree that thebrain, especially the frontal lobe, continuesto develop well into teenhood and beyond,many scientists hesitate to weigh in on thelegal debate. Some, like Giedd, say the data “just aren’t there” for them to confi-dently testify to the moral or legal culpa-bility of adolescents in court. Neuroscien-tist Elizabeth Sowell of UCLA says thattoo little data exist to connect behavior tobrain structure, and imaging is far frombeing diagnostic. “We couldn’t do a scanon a kid and decide if they should be triedas an adult,” she says.

Harper says the reason for bringing in“the scientific and medical world is not topersuade the court but to inform thecourt.” Fassler, who staunchly opposes thejuvenile death penalty, doesn’t want to pre-dict how the case will turn out. “It will beclose. I’m hopeful that the court will care-fully review the scientific data and willagree with the conclusion that adolescentsfunction in fundamentally different waysthan adults.” And perhaps, advocates hope,toppling the death penalty with a scientificunderstanding of teenagers will spread tobetter ways of rehabilitating such youths.

–MARY BECKMAN

Mary Beckman is a writer in southeastern Idaho.


CR

ED

IT:C

OR

BIS

Adolescence: Akin toMental Retardation?The human brain took center stage

in 2002 when the U.S. Supreme

Court ruled against the death

penalty for mentally retarded per-

sons. In that case (Atkins v.

Virginia), six of the nine justices

agreed that executing a convict

with limited intellectual capacity,

Daryl Atkins, would amount to cru-

el and unusual punishment. In-

structing the state of Virginia to

forgo the death penalty in such

cases, Justice John Paul Stevens

wrote: “Because of their disabilities

in areas of reasoning, judgment,

and control of their impulses,

[mentally retarded persons] do not act with the level of moral culpability that character-

izes the most serious adult criminal conduct.”

When the case of Christopher Simmons, who committed murder at age 17, comes be-

fore the same justices in October, says law professor Steven Drizin of Northwestern Uni-

versity in Chicago, defense attorneys hope to equate juvenile culpability to that of men-

tally retarded persons. “Juveniles function very much like the mentally retarded. The

biggest similarity is their cognitive deficit. [Teens] may be highly functioning, but that

doesn’t make them capable of making good decisions,” he says. Brain and behavior re-

search supports that contention, argues Drizin, who represents the Children and Family

Justice Center at Northwestern on the amicus curiae brief for Simmons. The “standard of

decency” today is that teens do not deserve the same extreme punishment as adults.

The Atkins decision provides advocates with a “template” for what factors should be

laid out to determine “evolving standards of decency,” says Drizin. These factors in-

clude the movement of state legislatures to raise the age limit for the death penalty to

18, jury verdicts of juvenile offenders, the international consensus on the issue, and

public opinion polls. In 2002, the court also considered the opinions of professional or-

ganizations with pertinent knowledge, which is how the brain research comes into play.

Last, the justices considered evidence that the mentally retarded may be more likely to

falsely confess and be wrongly convicted—a problem that adolescents have as well.

–M.B.

Last stop. In 2002, the Supreme Court rejected the

death penalty (6–3) for mentally retarded persons.

N E W S F O C U S

612

One of the most important legacies ofthe modern synthesis is the articula-tion of the biological species concept

(BSC). In his seminal 1942 work, System-atics and the Origin of Species (1), ErnstMayr defined species as “groups of actuallyor potentially interbreeding natural popula-tions, which are reproductively isolated from

other such groups.”Explicitly relating thedefinition of species tothe process of specia-tion, the BSC hasthrived—despite numer-ous hopeful alterna-tives—by inspiring awealth of literature onreproductive isolationand gene flow. The lasttwo decades in particular

have brought major advances in moleculargenetics, comparative analysis, mathemati-cal theory, and molecular phylogenetics;speciation has consequently matured froma field fraught with untestable ideas to onereaching clear, well-supported conclusions.

Jerry Coyne and Allen Orr’s Speciationprovides a much-needed review of these de-velopments. The exceedingly well-writtenand persuasive text eschews speculation. Theauthors instead resolutely develop testablecriteria for distinguishing alternative hy-potheses about evolutionary processes thatmay result in similar biological patterns, crit-ically evaluate how theoretical and empiricalresults meet the burden of proof, and active-ly confront important caveats and unresolvedquestions with practical suggestions. It is atestament both to the authors and to the stateof the field that the book provides such a ro-bust picture of the origin of species.

The “species problem” that Coyne andOrr consider in the book is how do speciesarise. More specifically, why do sexually re-producing organisms fall into discrete clus-ters? With this question in mind, they choosea relaxed version of the BSC that allows forsome gene flow among species as long asdistinctiveness is maintained. AlthoughCoyne and Orr recognize that the segrega-tion of biological diversity into distinct clus-

ters likely has multiple causes includingecology and history, the book concentratesalmost exclusively on reproductive isola-tion. Given that the bulk of speciation re-search has focused on the origin of repro-ductive barriers, the authors’ predominantfocus is, if nothing else, appropriate for theirchosen task. And although the book’s sub-ject matter and neontological approach mayprove unhelpful to systematists or paleontol-ogists, such readers may find some appease-ment in the favorable treatment of speciesselection and the thorough yet concise ap-pendix that discusses the relative merits andpitfalls of several major species concepts.

Proceeding from their premise thatstudying speciation is largely synonymouswith studying reproductive isolation, Coyneand Orr explore what we know about where,when, and how isolating barriers evolve.Following Mayr, they argue that speciationmost often occurs where populations are ge-ographically isolated or “allopatric.” The

broad range of theoretical conditions underwhich reproductive isolation evolves in al-lopatry, experimental evolution of reproduc-tive barriers in isolated laboratory popula-tions, and abundant examples of speciationevents associated with vicariance events orisolation on islands all strongly support thisposition. However, unlike Mayr, Coyne andOrr reach a more favorable though still un-enthusiastic view of sympatric speciation,one largely based on the development oftheoretical models with increasingly realis-tic assumptions that indicate sympatric spe-ciation could occur. They find empirical da-ta to be less compelling: only three casestudies not involving polyploid or hybridspeciation meet their criteria for a biogeo-graphic and evolutionary history that makesan allopatric phase highly unlikely.

An examination of when and how isolat-ing barriers evolve forms the core of Specia-tion. Far from being a dry catalog of mecha-nisms and well-known examples, these chap-ters offer engaging discussions that aim tosharpen how we define, detect, and measureisolating barriers; challenge us to decipherthe evolutionary rather than current impor-tance of these barriers; and synthesize evi-dence regarding their genetics and evolution.The treatment of mechanical isolation is anoften-entertaining read, and the overall at-tempt to outline and then disentangle hownatural and sexual selection may act to pro-mote nonecological or behavioral forms ofisolation is methodical and enlightening.

Speciation convincingly presents evi-dence for several once-unpopular theoriesthat have returned to dominate currentthinking. Most important among these is theprimacy of natural and sexual selection overdrift in driving speciation. Signatures ofpositive selection on genes involved inpostzygotic isolation and reproductive pro-teins as well as experimental evidence fromboth the lab and field connect adaptationand sexual selection to reproductive isola-tion. Another major finding is the congru-ence of the Dobzhansky-Muller model forthe evolution of postzygotic isolation withthe genetics of hybrid incompatibilities inmany natural systems. In contrast, classicalmodels of chromosomal speciation remainunpopular. Instead, chromosomal rearrange-ments are now cast as facilitators, ratherthan causal agents, of reproductive isolationbecause reduced recombination within theseregions restricts gene flow, thereby enablingthe accumulation of selected differences andhybrid incompatibilities.

The authors take cautious views on con-troversial questions like reinforcement,sympatric speciation, and diploid hybrid(or “recombinational”) speciation. For al- C

RED

IT:P

AIN

TIN

GS B

Y K

AK

O M

OR

ITA

,FR

OM

PLA

TES 4

6,4

7 O

F (5

)/C

OU

RT

ESY

SIN

AU

ER

The reviewers are in the Department of Biology,Indiana University, Bloomington, IN 47405, USA.E-mail: [email protected] (B.K.B.) [email protected] (L.H.R.)

E V O L U T I O N

How Species AriseBenjamin K. Blackman and Loren H. Rieseberg

Speciationby Jerry A. Coyne

and H. Allen Orr

Sinauer, Sunderland,

MA, 2004. 557 pp.

$89.95. ISBN 0-

87893-091-4. Paper,

$54.95, £34.99. ISBN

0-87893-089-2.

BOOKS et al.

Evidence for allopatry. These three congener-

ic Pacific wrasse (top to bottom: Halichoerestrimaculatus, H. margaritaceus, and H. hortu-lanus) were painted for David Starr Jordan and

Alvin Seale’s The Fishes of Samoa (5) by the

Japanese artist Kako Morita. (Morita’s work was

published with the help of Theodore Roosevelt,

who interceded after a government committee

ruled the plates were too expensive to print.)

Jordan argued that geographical barriers were

required for speciation.

30 JULY 2004 VOL 305 SCIENCE www.sciencemag.org

though recent theoretical advances demon-strate each phenomenon can occur under anontrivial set of conditions, conclusive em-pirical evidence that they occur in natureonly exists for the third process. Even so,the authors believe comparative analysesand further case studies will prove fruitfulavenues for determining if and how oftenthese processes operate in nature.

Coyne and Orr, who are Drosophila pop-ulation geneticists at the University ofChicago and the University of Rochester, re-spectively, provide remarkably lucid explana-tions of speciation phenomena in othergroups of organisms, alleviating prepublica-tion fears that the book would be dominatedby flies. Their discussion of polyploidy, forexample, is perhaps the best review of a pre-dominantly botanical literature by zoologists.Treatments of other plant-related topics likemating system isolation or hybridization areinsightful as well, but may raise eyebrows.For instance, unlike most botanical discus-sions of mating system evolution (2), Coyneand Orr argue that the shift from outbreedingto selfing is not a kind of reproductive isola-tion because gene flow is reduced as muchwithin as among taxa. Likewise, botanistsmay find an otherwise excellent treatment ofrecombinational speciation to be tilted towardthe evolution of postzygotic barriers throughhybridization as opposed to the contribu-tion of new hybrid gene combinations toecological differentiation and species es-tablishment. Lastly, the authors downplayincreasingly widespread phylogeneticevidence of cryptic introgression or hy-brid speciation in the plant, and now evenanimal, literature.

The book is a rich and thorough re-view, critique, and synthesis of recent lit-erature that is sure to become a classic readfor anyone interested in speciation. As theauthors’ purpose is to reflect on the value ofvarious approaches to evolutionary questionsand point out areas ripe for further investiga-tion, Speciation is not a textbook that pausesto give broad introductions; many methodsand terms are referred to in passing well be-fore being defined in later chapters. Despitethis, Coyne and Orr’s descriptions and logicalevaluations of theoretical and empirical workare remarkably clear and straightforward, aconsiderable achievement because the bookcovers material from complicated mathemat-ics to rigorous molecular genetics. An excel-lent book for a graduate seminar, Speciationshould also be interesting and accessible toscientists from diverse backgrounds.

Notably, many important results thatsupport Coyne and Orr’s conclusions in thebook have only been published in the lastyear. For instance, two of the four genesknown to underlie hybrid incompatibilitieswere identified only recently, and their

analysis adds great support to the role ofselection over drift in the evolution of thesebarriers (3, 4). With such research ongoing,and now with Speciation as a guide, the au-thors’ wish that their book “will stimulateyounger scientists to pursue their own workon speciation” will certainly be fulfilled.

References1. E. Mayr, Systematics and the Origin of Species

(Columbia Univ. Press, NY, 1942).2. D. A. Levin, Evol. Biol. 11, 185 (1978).3. D. A. Barbash, D. F. Siino, A. M. Tarone, J. Roote, Proc.

Natl. Acad. Sci. U.S.A. 100, 5302 (2003).4. D. C. Presgraves, L. Balagopalan, S. M. Abmayr, H. A.

Orr, Nature 423, 715 (2003).5. D. S. Jordan, A. Seale, Bull. U.S. Bur. Fish. 25, 173

(1906).

E V O L U T I O N

Hunting for OriginsR. Andrew Cameron

The term “Cambrian explosion” is real-ly a metaphor because the phenome-non named here is neither an explosion

nor did it happen in the Cambrian. Yes, thereappeared in Cambrian rocks (Chengjiangformation) dated 520 million years ago rep-resentatives of almost all major groups ofanimals. But newly estimated rates of

change in protein andDNA sequences calibrat-ed to well-dated fossils setthe divergences of thesemajor groups to a timewell before the Cambrian(1). Given this apparentcontradiction, many whostudy animal evolutionreckon that the early ani-

mals in these lineages were small and soft-bodied, resulting in a poor fossil record.Perhaps the conditions of the Cambrian en-vironment allowed the rapid appearance ofhard skeletal parts, greatly favored fossiliza-tion, or both.

In this context James Valentine (an emeri-tus professor of integrative biology at theUniversity of California, Berkeley) delivers anew book aimed at explaining the origin ofthe highest taxonomic groups of metazoans,On the Origin of Phyla. Considering the greatvariety of existing animals and the explana-tions for their elaboration, this is no easy job.There has been a steady trickle of books,some best sellers, offered to incorporateDarwinian evolution into a synthesis explain-ing the origin of higher taxa, but none havecome to represent the field the way that

The reviewer is in the Division of Biology, Mail Code139-74, 1200 East California Boulevard, CaliforniaInstitute of Technology, Pasadena, CA 91125, USA.E-mail: [email protected]

On the Origin

of Phylaby James W. Valentine

University of Chicago

Press, Chicago, 2004.

638 pp. $55, £38.50.

ISBN 0-226-84548-6.


B O O K S E T A L .

614

Darwin’s The Origin of Species has. Perhapsthis is due in part to the modern fragmenta-tion of studies in biology and geology intospecialized areas, which must be integrated tobuild an explanation, for as Valentine pointsout, ecologists have sought environmental ex-planations, developmental biologists have of-fered mechanisms for achieving variety ofform, and so on. But there has recently beena convergence of new data from several areasof research that is especially relevant to thesequestions. Witness the new finds of fossilpre-Cambrian embryos from the DoushantuoFormation of southwest China, estimated tobe 40 to 55 million years older than the baseof the Cambrian (2); the maturing consensusemerging from molecular phylogenetics; andadvances in the comparative molecular biolo-gy of development. Valentine has organizedthe book into sections that reflect these con-verging areas of study.

The question of when and how highertaxonomic groups like phyla evolved differsmarkedly from the one Darwin addressed145 years ago in The Origin of Species. It isnot simply different in scale but also in qual-ity. Although it is somewhat easier to seehow changes in single genes can lead to dif-ferences among species that render somemore capable of surviving in particular envi-ronments, it is more difficult to account forthe many changes that lead to entirely dif-ferent bodyplans as a simple accumulationof single-gene effects. For example, marinestickleback fishes possess bony plates andspines that presumably prevent predation,while their freshwater relatives show a lossof this armor through changes that can be at-tributed to a single gene (3, 4). However, en-

tire organ systems or embryonic germ lay-ers, features that distinguish higher taxa, canbe explained in terms of the gene regulatorynetworks whose architecture is hardwiredinto the genome. The simplest of these net-works leads to the specification of differen-tiated cell types, something like the cellmorphotypes that Valentine defines.

Networks that underlie morphogeneticpattern formation programs defining clade-specific body parts and bodyplans are morecomplex (5). These networks are assembledfrom cis-regulatory interactions that operateat the genome level. They represent the heri-table process by which the genome specifiesthe organism. From the structure of thesenetworks emerge obvious ways that majorchanges can happen through changes in thelinks among regulatory genes like transcrip-tion factors. For instance, a morphogeneticprogram may evolve with relatively minimalchanges to establish a new spatial domain ofexpression for a cell-differentiation program,and the resultant animal has a new body part.Although Valentine skirts around this mech-anistic model for the evolution of develop-mental programs in his definitions of the hierarchy of genomes, genes, and their pos-sible sources of change, he does not incorpo-rate a molecular model in his final synthesis.

The intellectually greedy might flip tothe back of the book, skipping over the sec-tions on the evidence for the origins of meta-zoan phyla and the descriptions of the phylathemselves, even though these compilations(particularly the section on the fossil record)are the work’s strong points. In the third sec-tion, Valentine paints his view of the evolu-tion of the phyla, with an emphasis on the

work of morphologists and paleontologists(a portion of which he contributed). Hefleshes out a scenario of variation and ex-tinction that unites the detailed descriptionsof the first two sections. The pictures he de-lineates here reveal correlations uniting dif-ferent levels of biological organization, butabsent are firm statements about causalmechanisms from which predictions couldbe made. Only in the section’s closing pagesare genomes considered.

In view of the volatility of the ideas andthe controversy that still exist in this partic-ular area of evolutionary biology, onemight argue that it is too early to explainthe causes of the origin of phyla. But asValentine aptly points out, the time willnever be exactly right: there are alwaysmore information to incorporate and moreideas to organize. Though too heavy withdata to be carried in the kit bag, On theOrigin of Phyla is a likely candidate for thebookshelves of those who hunt for pre-Cambrian fossils or the historical patternsin DNA sequences.

References and Notes1. K. J. Peterson et al., Proc. Natl. Acad. Sci. U.S.A. 101,

6536 (2004).2. J.-Y. Chen et al., Science 305, 218 (2004); published

online 3 June 2004 (10.1126/science.1099213).3. W. A. Cresko et al., Proc. Natl. Acad. Sci. U.S.A. 101,

6050 (2004).4. J. S. McKinnon et al., Nature 429, 294 (2004).5. E. H. Davidson et al., Science 295, 1669 (2002).6. This reconstruction is from the Smithsonian

Institution’s Traveling Exhibition The Burgess Shale:Evolution’s Big Bang, at the Sternberg Museum ofNatural History, Hays, KS, through 24 October; theBurke Museum of Natural History and Culture,Seattle, WA, 13 November 2004 to 1 May 2005; andthe Sam Noble Oklahoma Museum of NaturalHistory, Norman, OK, 21 May to 27 November 2005. C

RED

IT:M

AR

Y P

AR

RIS

H/D

EPT.O

F PA

LEO

BIO

LOG

Y,N

AT

ION

AL

MU

SEU

M O

F N

AT

UR

AL

HIS

TO

RY

,SM

ITH

SO

NIA

N I

NST

ITU

TIO

N

An odd ecdysozoan. Phylogenetic analyses using morphological traits

place Opabinia and the anomalocarids in a clade basal to Arthropoda or

Onychophora. But an alternative interpretation of the imputed lobopodi-

al structures in this predator from the Burgess Shale suggests it is a stem-

group arthropod. (6)


B O O K S E T A L .

616

As a signatory to the United NationsFramework Convention on ClimateChange (UNFCCC), the United

States shares with many countries its ulti-mate objective: stabilization of greenhousegas concentrations in the atmosphere at alevel that prevents dangerous interferencewith the climate system. Meeting this UNFCCC objective will require a long-termcommitment and international collaboration.

President Bush’s policy on climatechange harnesses the power of markets andtechnological innovation, maintains eco-nomic growth, and encourages global par-ticipation. Although climate change is acomplex and long-term challenge, theBush administration recognizes that thereare cost-effective steps we can take now.

Near-Term Policies and MeasuresIn 2002, President Bush set a national goalto reduce the greenhouse gas intensity (1)of the U.S. economy by 18% by 2012. Thisgoal sets America on a path to slow thegrowth in greenhouse gas emissions and—as the science justifies and the technologyallows—to stop and reverse that growth asneeded to meet the UNFCCC goal (2). Ourapproach focuses on reducing emissionswhile sustaining the economic growthneeded to finance investment in new, cleanenergy technologies. The administrationestimates that this commitment willachieve about 100 million metric tons ofcarbon equivalent (MMTCe) of reducedemissions in 2012, with more than 500MMTCe in cumulative savings over thedecade (3).

To this end, the administration has devel-oped an array of policy measures, includingfinancial incentives and voluntary programs.For example, our Climate VISION (4),Climate Leaders (5), and SmartWayTransport Partnership (6) programs workwith industry for voluntary reduction ofemissions. The Department of Agriculture isusing its conservation programs to providean incentive for actions that increase carbon

sequestration (7). We also are pursuing manyenergy supply technologies with compara-tively low or zero CO2 emissions profiles,such as solar, wind, geothermal, bioenergy,and combined heat and power. The presidenthas proposed more than $4 billion in taxcredits as incentives for these and other en-ergy-efficient technologies over the next 5years (3). Last year, the Bush administrationincreased fuel economy standards for newlight trucks and sport utility vehicles by 1.5miles per gallon over the next three modelyears, leading to the estimated avoidance of9.4 MMTCe of emissions (8).

While acting to slow the pace of green-house gas emissions in the near term, theUnited States is laying a strong scientificand technological foundation to reduce un-certainties, to clarify risks and benefits,and to develop realistic mitigation optionsto meet the UNFCCC objective.

Advancing Climate Change ScienceIn 2001, President Bush commissioned theNational Research Council (NRC) to ex-amine the state of our knowledge and un-derstanding of climate change science. TheNRC’s report (9) makes clear that there arestill important gaps in our ability to meas-ure the impacts of greenhouse gases on theclimate system. Major advances in under-standing and modeling of the factors thatinfluence atmospheric concentrations ofgreenhouse gases and aerosols, as well asthe feedbacks that govern climate sensitiv-ity, are needed to predict future climatechange with greater confidence.

Last summer, the Climate ChangeScience Program (CCSP) released a newstrategic plan that addresses these gaps(10). The plan is organized around fivegoals: (i) improving our knowledge of cli-mate history and variability; (ii) improvingour ability to quantify factors that affectclimate; (iii) reducing uncertainty in cli-mate projections; (iv) improving our un-derstanding of the sensitivity and adapt-ability of ecosystems and human systemsto climate change; and (v) exploring op-tions to manage risks. Annually, almost $2billion is spent on climate change scienceby the federal government.

A review of the CCSP plan by NRCshows the administration is on the righttrack. While concern was expressed aboutfuture funding to execute the plan, theNRC concluded that it “articulates a guid-ing vision, is appropriately ambitious, andis broad in scope” (11).

NRC’s report also identified the realneed for a broad global observation systemto support measurements of climate vari-ables. Last June, the United States hostedmore than 30 nations at the inaugural EarthObservation Summit, out of which came acommitment to establish an intergovern-mental, comprehensive, coordinated, andsustained Earth observation system. Thedata collected by the system will be used tocreate better climate models, to improveour knowledge of the behavior of CO2 andaerosols in the atmosphere, and to developstrategies for carbon sequestration.

Accelerating Climate ChangeTechnology DevelopmentThe Bush administration also is movingahead on advanced technology options thathave the potential to substantially reduce,avoid, or sequester future greenhouse gasemissions. About 80% of current green-house gas emissions are energy related,and, although projections vary consider-ably, a tripling of energy demand by 2100is not unimaginable (12). Therefore, to pro-vide the energy necessary for continuedeconomic growth while we reduce green-house gas emissions, we may have to de-velop and deploy cost-effective technolo-gies that alter the way we produce and useenergy.

By 2100, more than half of the world’senergy may have to come from low- or zero-emission technologies to attain the UNFCCC goal (13). The pace and scope ofneeded change will be driven partially byfuture trends in greenhouse gas emissionsthat, like climate sensitivity, are uncertain.The complex relations among populationgrowth; economic development; energydemand, mix, and intensity; resource avail-ability; technology; and other variablesmake it impossible to accurately predict fu-ture greenhouse gas emissions on a 100-year time scale.

The Climate Change TechnologyProgram (CCTP) was created to coordinateand prioritize the federal government’snearly $3 billion annual investment in cli-mate-related technology research, develop-ment, demonstration, and deployment(RDD&D). Using various analytical tools,CCTP is assessing different technology op-tions and their potential contributions to

C L I M AT E

The Bush Administration’s

Approach to Climate Change Spencer Abraham

POLICY FORUM

The author is the U.S. Secretary of Energy, 1000Independence Avenue, SW, Washington, DC 20585,USA.


617

reducing greenhouse gasemissions. Given the tremen-dous capital investment in ex-isting energy systems, the de-sired transformation of ourglobal energy system maytake decades or more to im-plement fully. A robustRDD&D effort can make ad-vanced technologies availablesooner rather than later andcan accelerate modernizationof capital stock at lower costand with greater flexibility.

CCTP’s strategic vision hassix complementary goals: (i)reducing emissions from ener-gy use and infrastructure; (ii)reducing emissions from ener-gy supply; (iii) capturing andsequestering CO2; (iv) reduc-ing emissions of other green-house gases; (v) measuring and monitoringemissions; and (vi) bolstering the contribu-tions of basic science (14).

Ten federal agencies support a portfolioof activities within this framework.Annually, more than $700 million is beingspent to advance energy efficiency tech-nologies (plus $500 million for accelerateddeployment), and more than $200 millionsupports renewable energy. Many activitiesbuild on existing work, but the Bush ad-ministration also has expanded and re-aligned some activities and launched newinitiatives in key technology areas to sup-port the CCTP’s goals.

In his 2003 State of the Union address,President Bush made a commitment to thedevelopment of a hydrogen economy,pledging $1.7 billion over 5 years for hisHydrogen Fuel Initiative and Freedom-CAR Partnership to develop hydrogen fuelcell–powered vehicles. The transition tohydrogen as a major energy carrier over thenext few decades could transform the na-tion’s energy system and create opportuni-ties to increase energy security by makingbetter use of diverse domestic energysources for hydrogen production and to re-duce emissions of air pollutants and CO2(15). Where hydrogen is produced fromfossil fuels, we must also address carboncapture and sequestration.

To help coordinate and leverage ongo-ing work overseas, the United States led theeffort to form the International Partnershipfor the Hydrogen Economy (IPHE). IPHEwill address the technological, financial,and institutional barriers to hydrogen andwill develop internationally recognizedstandards to speed market penetration ofthe new technologies.

The administration also is pursuingnext-generation nuclear energy as a zero-

emissions energy supply choice. TheGeneration IV International Forum, withnine other nations as partners, is workingon reactor designs that are safe, economi-cal, secure, and able to produce new prod-ucts, such as hydrogen. Six promisingtechnologies have been selected as candi-dates for future designs and could beready as early as 2015. In 2003, PresidentBush announced that the United Stateswould join the ITER project to developfusion as an energy source. Although thetechnical hurdles are substantial, thepromise of fusion is simply too great toignore.

Carbon capture and sequestration is acentral element of CCTP’s strategy becausefor the foreseeable future, fossil fuels willcontinue to be the world’s most reliable andlowest-cost form of energy. It is unrealisticto expect countries—particularly develop-ing countries—with large fossil reserves toforgo their use. A realistic approach is tofind ways to capture and store the CO2 pro-duced when these fuels are used.

The Department of Energy is currentlyworking on 65 carbon sequestration proj-ects around the country. In the last 2 years,we have increased the budget for these ac-tivities 23% to $49 million. The multilater-al Carbon Sequestration LeadershipForum, a presidential initiative inauguratedin June 2003 with 16 partners, will set aframework for international collaborationon sequestration technologies.

The forum’s partners are eligible to par-ticipate in FutureGen, a 10-year, $1 billiongovernment-industry effort to design,build, and operate the world’s first emis-sions-free coal-fired power plant. Thisproject, which cuts across many CCTPstrategic goals, will employ the latest tech-nologies to generate electricity, produce

hydrogen, and sequester CO2 from coal.Through this research, clean coal can re-main part of a diverse, secure energy port-folio well into the future.

These initiatives and other technologiesin the CCTP portfolio (16) could revolu-tionize energy systems and put us on a pathto ensuring access to clean, affordable en-ergy supplies while dramatically reducinggreenhouse gas emissions. The figure, left,offers a glimpse of the range of emissionsreductions new technologies might makepossible in energy end use, energy supply,carbon sequestration, and other greenhousegases on a 100-year scale and across arange of uncertainties.

The Bush administration has developeda comprehensive strategy on climatechange that is informed by science, empha-sizes innovation and technological solu-tions, and promotes international collabo-ration to support the UNFCCC objective.Although the scientific and technologychallenges are considerable, the presidentremains committed to leading the way onclimate change at home and around theworld.

References and Notes1. Measured as the ratio of greenhouse gases (carbon

equivalent) emitted per real gross domestic product.

2. See www.whitehouse.gov/news/releases/2002/02/

addendum.pdf.

3. Global Climate Change Policy Book: A New Approach

(The White House, Washington, DC, 14 February

2002); available at www.whitehouse.gov/news/

releases/2002/02/climatechange.html.

4. See www.climatevision.gov.

5. See www.epa.gov/climateleaders.

6. See www.epa.gov/smartway.

7. See www.usda.gov/news/releases/2003/06/fs-0194.htm.

8. National Highway Traffic Safety Administration, Final

Environmental Assessment: National Highway Traffic

Safety Administration Corporate Average Fuel

Economy (CAFE) Standards (NHTSA,Washington, DC,

2003); available at: www.nhtsa.dot.gov/cars/rules/

cafe/docs/239533_web.pdf.

9. National Research Council, Climate Change Science:

An Analysis of Some Key Questions, Committee on

the Science of Climate Change (National Academy

Press, Washington, DC, 2001), pp. 20–21.

10. CCSP, Strategic Plan for the U.S. Climate Change

Science Program (CCSP, Washington, DC, July 2003);

available at www.climatescience.gov.

11. National Research Council, Implementing Climate

and Global Change Research: A Review of the Final

U.S. Climate Change Science Program Strategic Plan

(National Academies Press, Washington, DC, 2004),

p. 1.

12. Intergovernmental Panel on Climate Change, “An

overview of the scenario literature,” Emissions

Scenarios (Cambridge Univ. Press, Cambridge, 2000).

13. See, for example, K. Caldeira, A. K. Jain, M. I. Hoffert,

Science 299, 2052 (2003).

14. CCTP, U.S. Climate Change Technology Program Draft

Strategic Plan: Vision and Framework (CCTP,

Washington, DC, in preparation); see www.

climatetechnology.gov.

15. National Research Council, The Hydrogen Economy:

Opportunities, Costs, Barriers, and R&D Needs

(National Academies, Washington, DC, 2004).

16. CCTP, Research and Current Activities (CCTP,

Washington, DC, 2003); available at www.

climatetechnology.gov.

400

350

300

250

200

150

100

50

0Energyend-use

Energysupply

Sequestration Othergreenhouse

gases

Scenario 1Scenario 2Scenario 3

Gig

ato

nn

es o

f ca

rbo

n e

qu

ival

ent

Potential contributions to emissions reduction

Potential ranges of greenhouse gas emissions reductions to

2100 by category of activity for three technology scenarios

characterized by viable carbon sequestration (scenario 1); dra-

matically expanded nuclear and renewable energy (scenario 2);

and novel and advanced technologies (scenario 3) (14).

P O L I C Y F O R U M


618

Understanding how individual traitsof organisms affect both their inter-actions with other species and the

dynamics of the ecosystem community isan important challenge confronting evolu-tionary ecologists. A major problem hasbeen the lack of genotypes that differ ex-

clusively in the traitof interest as well asa lack of techniquesfor assessing ex-pression of such

genotypes. Recent breakthroughs in mo-lecular biology have provided ecologistswith exciting new tools with which to pur-sue an ecogenomics strategy (1, 2). Nowecologists can perform delicate geneticmanipulations to obtain well-characterizedgenotypes, which, together with mechanis-tic knowledge of phenotypic plasticity,provide information on the effects of indi-vidual traits on species interactions in com-munities. A major obstacle to the rapid in-corporation of molecular tools into ecolog-ical studies, however, is that most modelorganisms used by molecular biologists donot yet match the model organisms favoredby ecologists. This is changing, as exem-plified by the study of Kessler et al. (3)published on page 665 of this issue. Theseauthors incorporate molecular techniquesinto an ecological field study to investigatethe effects of three genes on the interac-tions between native tobacco plants andtheir natural insect herbivores.

For more than a decade, Baldwin andcolleagues have analyzed the responses ofnative tobacco plants induced by insectherbivores, with methods ranging fromtranscriptome analyses in the laboratory(1) to field studies of phenotypically ma-nipulated plants (4). They have developedthree tobacco plant lines in which theyhave genetically silenced one of threegenes (encoding lipoxygenase, hydroper-oxide lyase, or allene oxide synthase) ofthe oxylipin signaling pathway. Oxylipin

signaling mediates both direct (toxins) andindirect (natural enemies) plant defensesagainst herbivorous attack. Exposing thesethree tobacco plant genotypes togetherwith a wild-type control to a natural insectcommunity in Utah enabled the investiga-tors to elucidate the effects of the threegenes on host-plant selection by insect her-bivores (3).

Plant-insect interactions comprise a ma-jor biological interface in terrestrial ecosys-tems (5) (see the Perspective on page 619).The ecology of these interactions has beenthe subject of intensive studies not only be-cause such interactions are of fundamentalinterest, but also because they provide thebasis of ecologically sound management ofagricultural pests. Two major investigativethemes are the complex of direct and indi-rect defenses that plants employ against in-sects, and the adaptations of insects to these

plant defenses (5). Plant defenses may af-fect the performance of herbivorous insectsdirectly, or they may do so indirectly by en-hancing the effectiveness of the herbivores’natural enemies (see the figure). Both typesof defense may be either constitutive(switched on all the time) or may be in-duced only in response to herbivore attack(6).

In the new work, Kessler et al. (3) in-vestigate the phenotypic effects of silenc-ing genes known to be involved in directand indirect defenses against insect herbi-vores. Under field conditions and after si-lencing of lipoxygenase-dependent signal-ing, the authors reveal that the tobaccoplants not only were more susceptible toherbivory, but also showed a reduction intheir emission of plant volatile organiccompounds (which may affect attraction ofthe herbivore’s enemies) in response to her-bivore attack. The up-regulation of de-fense-related genes and the down-regula-tion of photosynthesis-related genes wasobserved in the wild-type control plantsbut not in the engineered plants. Kesslerand Baldwin (4) previously demonstratedthat induced tobacco volatiles attract pred-ators that remove herbivore eggs. Giventhat silencing of the lipoxygenase pathwayresulted in a reduction of volatile emission

E C O L O G Y

Ecogenomics Benefits

Community EcologyMarcel Dicke, Joop J. A. van Loon, Peter W. de Jong

PERSPECTIVES

The authors are in the Laboratory of Entomology,Wageningen University, Post Office Box 8031, NL-6700 EH Wageningen, Netherlands. E-mail:[email protected]

OH

A

Direct defense phenotype Indirect defense phenotypeExpressed plant genotype

D FE

B C

3 5 7 9 11 13 15 17 19 21 23

O

Beating back the bugs. A tritrophic system consisting of plants, insect herbivores, and their natural en-

emies.This particular system comprises cabbage plants (Brassica oleracea) (B), herbivorous larvae of the

cabbage white butterfly (Pieris brassicae) (A), and parasitoids, Cotesia glomerata (C), that attack P.brassicae caterpillars (C). Damage caused by caterpillars feeding on cabbage plants up-

regulates the expression of various genes in the plants, which are visualized as red spots in the mi-

croarray (E), and down-regulates the expression of other genes (green spots). Herbivory induces up-reg-

ulation of the biosynthesis of certain types of glucosinolates (D), toxic secondary metabolites charac-

teristic of the Brassicaceae that mediate a direct defense against herbivorous insects. Additionally, the

emission of dozens of volatile organic compounds, each represented by a peak in the gas chromatogram

(F), is induced by herbivory. These herbivore-induced volatiles act as an indirect defense by attracting

parasitoids that lay eggs in the caterpillars. Shown are the green-leaf volatile (Z)-3-hexen-1-ol and the

terpenoid 1,8-cineole, representatives of two dominant classes of volatiles emitted by cabbage (F). CR

ED

IT:H

AN

S M

.SM

ID A

ND

MA

NA

BU

KA

MIM

UR

A

Enhanced online at

www.sciencemag.org/cgi/

content/full/305/5684/618


619

(3), a logical follow-up experiment wouldbe to assess the effects of gene silencing onplant consumers at higher trophic levels.

Other studies have exploited transgenicplants in the laboratory to investigate the ef-fects of individual genes on ecological inter-actions, such as the costs of resistance to apathogen (direct defense) (7) or the attrac-tion of the enemies of herbivorous insects(indirect defense) (8). Such laboratory stud-ies are valuable for assessing the effects ofgenes on particular interactions. However,the ultimate assessment of a gene’s functionshould be made under field conditions, andit is here that Kessler et al. (3) make theirbiggest contribution to the study of directplant defense against herbivory.

Plants are endowed with a remarkablecapacity to compensate for herbivore dam-age (9); the relationship between plantdamage and seed production is complexand context dependent. The selective ad-vantage of induced plant defenses shouldbe detectable at the level of plant fitness,the ultimate evolutionary currency (see thePerspective on page 619). However, studiesof transgenic plants in the field are subjectto regulatory constraints—particularlywith regard to the dissemination of trans-genic plant pollen—which limits progressin this area.

Apart from exploiting the new opportu-nities that transgenic plants offer, ecolo-gists can take advantage of other moleculartools to make unprecedented progress inunderstanding the ecology of plant-insectinteractions. Adaptations of insects to plantdefenses may be spatially variable and may

lead to local variations in the genetics un-derlying such adaptations (10). Conse-quently, investigating the interaction be-tween selective regimes on the one handand migration, gene flow, and demographyon the other can answer important evolu-tionary questions—for example, how toexplain the present distribution of an adap-tation to a plant defense? This question canbe analyzed by integrating genetic data atneutral loci and loci under selection—anapproach called population genomics(11)—with ecological data and spatialmodeling. The population genomics ap-proach involves genome-wide sampling ofmarkers and comparing the variation in pu-tative neutral markers with that in markersthought to be associated with genomic re-gions under selection. This enables the dis-tinction between genome-wide effects (forexample, those caused by migration) andlocus-specific effects (for example, thoseinfluenced by selection). Such an approachwill lead to new insights into the spreadingof adaptive traits and the geographic distri-bution of these traits.

We have highlighted only two of themany molecular strategies that can beadopted by evolutionary ecologists. Byadopting such molecular strategies, ecolo-gists can study mechanism and function in-tegratively as both the expressed genotypeand the performance of the phenotype withother members of the community areknown. Molecular techniques provide evo-lutionary ecologists with many more oppor-tunities. As a representation of an organ-ism’s phenotype, microarrays may be used

to assess phenotypic variation under naturalconditions or after artificial selection.Molecular expression markers that havebeen selected on the basis of genomic infor-mation may be used to screen segregatingpopulations for genotypes that show quanti-tative variation in the expression of a geneof interest (12) rather than the qualitativedifference represented by a silenced indi-vidual gene and its wild-type control. Thisprovides an even better set of genotypes foraddressing questions about the evolution-ary ecology of plant-insect interactions.The important contribution of Kessler et al.(3) will likely mark the beginning of a bigleap forward in our comprehension of theecology of plant-insect interactions.

References and Notes1. A. Kessler, I. T. Baldwin, Annu. Rev. Plant Biol. 53, 299

(2002).2. M. Dicke, R. M. P. van Poecke, J. G. de Boer, Basic

Appl. Ecol. 4, 27 (2003).3. A. Kessler, R. Halitschke, I.T. Baldwin, Science 305, 665

(2004); published online 1 July 2004 (10.1126/science.1096931).

4. A. Kessler, I. T. Baldwin, Science 291, 2141 (2001).5. L. M. Schoonhoven, T. Jermy, J. J. A. van Loon, in

Insect-Plant Biology. From Physiology to Evolution(Chapman & Hall, London, 1998).

6. M. Dicke, J. J. A. van Loon, Entomol. Exp. Appl. 97, 237(2000).

7. J. Bergelson, C. B. Purrington, C. J. Palm, J. C. Lopez-Gutierrez, Proc. R. Soc. London B 263, 1659 (1996).

8. R. M. P. van Poecke, M. Dicke, J. Exp. Bot. 53, 1793(2002).

9. S. Y. Strauss, A. A. Agrawal, Trends Ecol. Evol. 14, 179(1999).

10. J. N. Thompson, Am. Nat. 153, S1 (1999).11. W. C. Black, C. F. Baer, M. F. Antolin, N. M. DuTeau,

Annu. Rev. Entomol. 46, 441 (2001).12. E. E. Schadt et al., Nature 422, 297 (2003).13. The authors are supported by a VICI grant from the

Netherlands Organisation for Scientific Research.

Ever since Bates, Darwin, and Wallace,ecologists and evolutionary biologistshave strived to uncover the causes of

high species diversity in tropical regions. Inthe case of plant diversity, early studies inthe tropical regions of Central and SouthAmerica emphasized the contribution of bi-otic interactions, particularly the effects ofplant-eating predators (herbivores) on plantsurvival. In contrast, efforts in tropical Asiahave centered on abiotic factors, such as soilnutrients and water, describing how thespecies composition of forests shifts withchanges in soil type. It is likely that neitherfactor works alone. Tropical forests grow on

a patchwork of different soil types such thatmany tree and plant species are uniquelyadapted to particular soil (edaphic) condi-tions. For example, on tropical white-sandsoils—which are extremely low in nutrientsand water-holding capacity and high inacidity and aluminum—the vegetation con-sists of many tree species that grow nowhereelse. In 1974, Janzen (1) proposed that themain factor constraining plants from grow-ing on tropical white-sand soils is attack byherbivorous predators rather than the inabil-ity of plants to sustain positive growth underlow nutrient conditions. On page 663 of thisissue, Fine et al. (2) report results of the firstreal field test of this hypothesis. Their find-ings provide evidence that herbivores influ-ence tropical forest diversity by contributingto habitat specialization (3).

One pathway by which herbivores influ-ence plant diversity is by decimating plantpopulations, particularly at the seed andseedling stage (4). This decimation is per-haps most common where many of the her-bivores are large, as in tropical Africa andthe Indian subcontinent. Mega-herbivoresin these regions—including elephants, gi-raffes, antelope, wildebeests, and rhinos—not only change the number and identity ofplant species but convert forest vegetationto grasslands (5).

Herbivores need not be large, however,to influence plant diversity. The secondpathway by which herbivores influencetropical plant diversity, and plant diversityin general, is by shifting the competitivebalance among plant species. If herbivores,including insects, preferentially attackseeds, seedlings, and juveniles of the mostcompetitive species or the most abundantspecies (frequency-dependent attack),poorer competitors or less frequent plantspecies could be maintained in the system.For example, the Janzen-Connell hypothe-

E C O L O G Y

Herbivores RuleRobert J. Marquis

The author is in the Department of Biology,University of Missouri–St. Louis, St. Louis, MO 63121,USA. E-mail: [email protected]


P E R S P E C T I V E S

620

sis (6, 7) contends that specialist herbivorespreferentially attack offspring that are closeto parent trees, enabling dispersed off-spring to escape. The influence of insectherbivores on the survival of seedlings (8)and patterns of density-dependent mortali-ty (9) in tropical forests support this hy-pothesis.

A third mechanism by which herbivoresmight promote tropical forest diversity isby causing or reinforcing habitat special-ization. This is the predicted outcome ofJanzen’s 1974 hypothesis (1). Nutrientavailability is so low on these nutrient-poorsoils that loss of plant tissue to herbivoreattack, and the attendant loss of nutrients,places a poorly defended plant genotype ata great disadvantage. Thus, there would bestrong selection for a robust, but costly, anti-herbivore defense in plants growing onnutrient-poor soils. In the presence of her-bivores, no species growing in surroundingforest regions on more nutrient-rich soils—clay soils in the Fine et al. study (2)—would invade white-sand soils becausethese species would not be sufficientlywell-defended to protect themselvesagainst nutrient loss associated with herbi-vore attack. In the absence of herbivores,however, these species could invade nutri-ent-poor white-sand soils because, withoutherbivores, the balance is tipped againstwhite-sand soil specialist species. Like-wise, white-sand soil specialists could notinvade clay soils because the cost of de-fense investment has been too high to allowthem to compete against species adapted tomore nutrient-rich soils.

How might one test this hypothesis? Anobvious experiment is to reciprocally trans-plant both kinds of plant species to bothsoil types and quantify their success in thepresence and absence of herbivores (see thefigure). Fine and colleagues (2) have com-pleted just such an experiment in thePeruvian Amazon. At the study site, theytransplanted 20 species of tree seedlingsfrom six genera, paired such that each paircomprised a clay soil specialist species anda white-sand soil specialist species. Theythen erected herbivore exclosures consist-ing of a fine mesh that excluded insect her-bivores from one-half of the seedlings ineach habitat.

The results of this experiment demon-strate that herbivores help to maintain habi-tat segregation of the different tree species.On white-sand soils in the absence of her-bivores, clay soil specialists survived betterand grew more than did white-sand soilspecialists. No longer burdened by the lossof resources to herbivores, they outpacedwhite-sand species whose growth was pre-sumably constrained by their higher invest-ment in antiherbivore defenses. In contrast,when herbivores were free to attack bothtypes of tree species growing on white-sand soils, the white-sand soil specialistsfared much better. These results suggestthat in the absence of herbivores, clay soilspecialists could invade white-sand soils,perhaps even displacing white-sand soilspecialists. Meanwhile, on clay soils,white-sand soil species did not grow aswell as did clay soil species, again presum-ably because they were constrained by their

high investment in defense against her-bivory. Thus, a large allocation of resourcesto defense represents a burden to white-sand soil species when on clay soils, but anecessary evil when they are on white-sandsoils inhabited by herbivores.

Taking an evolutionary perspective re-garding these biotic interactions raises thequestion of whether herbivores act not on-ly as maintainers but also as promoters ofhigh species diversity. Fine et al. (2) sug-gest that herbivores may have promoteddiversification through parapatric specia-tion (where new species form across envi-ronmental discontinuities) by accentuatingselective gradients among habitats. Here,new species would form on white-sandsoils as a result of strong selection by her-bivore-soil interactions in the face of con-tinuous gene flow. By determining theevolutionary tree for clay soil and white-sand soil specialists, one could deducewhich traits evolved first as white-sandsoils were colonized. For example, did de-fense traits evolve first or did those moredirectly related to plant-soil relationships,such as special morphologies of plantroots or traits related to interactions withfungi in mycorrhizae?

The Fine et al. study provides strongsupport for the idea that herbivores con-tribute to the maintenance of high speciesdiversity in local areas that vary in soiltype. Herbivores are probably not the solefactor, however. Efforts such as the Fine etal. study suggest the need for a holistic ap-proach to solving questions relating tocommunity ecology. A parochial focus on C

RED

IT:T

AIN

A L

ITW

AK

XX

Herbivore exclosure Control (roof only)

White-sand soil Lateritic red-clay soil

Reciprocal transplant of tropical tree seedlings. Twenty species of

tropical tree seedlings (two species are shown) were reciprocally trans-

planted onto white-sand soils and lateritic red-clay soils. One-half of the

tree species used in this experiment naturally grow only on clay soils,

whereas the other half are white-sand soil specialists (2). Exclosures,

which prevented insect herbivores from gaining access to the seedlings,

were erected around one-half of the seedlings of each species planted on

each soil type. A control exclosure (roof only) was erected around the re-

mainder of the transplanted seedlings, which thus were vulnerable to at-

tack by insect herbivores. Seedling survival and growth were monitored in

response to soil source type, soil type at the site of transplanting, and ex-

posure to herbivores.



621

abiotic or biotic factors alone is likely toprovide only limited answers. In addition tosoils, the impact of herbivores on tropicalforests may vary with elevation and alonggradients or discontinuities in soil flooding(10), light (11), and fire (12). But for now,the Fine et al. work adds to the mountingevidence that herbivory is a major factordetermining the plant composition of trop-ical forests.

References1. D. H. Janzen, Biotropica 6, 69 (1974).

2. P. V. A. Fine, I. Mesones, P. D. Coley, Science 305, 663

(2004).

3. R. J. Marquis, in Biotic Interactions in the Tropics, D.

Burslem, M. Pinard, S. Hartley, Eds. (Cambridge Univ.

Press, Cambridge, in press).

4. S. M. Louda, Ecol. Monogr. 52, 25 (1982).

5. A. R. E. Sinclair, P. Arecese, Eds., Serengeti II,

Dynamics, Management, and Conservation of an

Ecosystem (Univ. of Chicago Press, Chicago, IL, 1995).

6. D. H. Janzen, Am. Nat. 104, 501 (1970).

7. J. H. Connell, in Dynamics of Populations, P. J. den

Boer, G. R. Gradwell, Eds. (Centre for Agricultural

Publishing and Documentation, Wageningen, Neth-

erlands, 1971), pp. 298–312.

8. L. A. Hyatt et al., Oikos 103, 590 (2003).

9. K. E. Harms et al., Nature 404, 493 (2000).

10. R. T. King, Biotropica 35, 462 (2003).

11. S. J. DeWalt et al., Ecology 85, 471 (2004).

12. H. T. Dublin, in Serengeti II, Dynamics, Management,

and Conservation of an Ecosystem, A. R. E. Sinclair,

P. Arecese, Eds. (Univ. of Chicago Press, Chicago,

1995), pp. 71–90.

The geologic time scale stands as a ma-jor achievement of 19th-century sci-ence, a coherent record of our planet’s

history fashioned from myriad details of in-dividual rock outcroppings. The eras, peri-ods, and finer divisions of the scale not onlycodify geologic time, they reflect our accu-mulated understanding of Earth’s past—or atleast its more recent past. The CambrianPeriod, with its fossil record of animal diver-sification, began only 543 million years ago(Ma), when Earth was already 4000 millionyears old (see the figure). In the 19th centu-ry and for much of the 20th century, the be-ginning of the Cambrian (also the beginningof the Paleozoic era and the Phanerozoiceon) marked the most distant temporalreaches of Earth’s tractable historical record.The absence of skeletonized fossils thatmark Phanerozoic time made Precambrianrocks difficult to correlate, and so the finestratigraphic divisions of the younger recordgave way to broad intervals that permittedonly limited insight into foundational eventsof Earth history. In 1991, perhaps out of res-ignation, the International Union of Geo-logical Sciences (IUGS) approved a divisionof Precambrian time into eons, eras, and pe-riods defined strictly by chronometric age,without reference to events recorded in sedi-mentary rocks (1). The eras stuck, but theproposed period names are seldom used.

This tradition was swept aside in Marchthis year with the approval by IUGS of anaddition to the geologic time scale: theEdiacaran Period (2). This newly ratifiedperiod, which directly precedes theCambrian, is the first Precambrian intervalto be defined according to the principlesthat govern the Phanerozoic time scale. It isalso the first stratigraphically defined newperiod of any sort to be added since 1891when Williams divided the CarboniferousPeriod in two (Mississippian and Penn-sylvanian). The distinctive character of theEdiacaran interval has been recognized fordecades, and numerous geologists—includ-

ing Sokolov, Termier and Termier, andCloud and Glaessner (2)—have proposedformal definitions of this interval. Now, inaccordance with international rules, thenew period has been defined by an eventrecorded in a single section of rock out-cropping termed the global stratotype sec-tion and point (GSSP). (The GSSP is thereference section that defines the “stan-dard” for recognition of the base of the newperiod worldwide.) The initial GSSP of theEdiacaran Period lies at the base of a textu-rally and chemically distinctive carbonatelayer that overlies glaciogenic rocks in anexposure along Enorama Creek in theFlinders Ranges, South Australia (2) (seethe figure). The period’s end coincides withthe beginning of the Cambrian Period,which is defined by its own initial GSSP re-siding in Newfoundland, Canada.

Formalisms aside, international ratifica-tion of the new period reflects our expand-ing knowledge of Earth’s deep physical andbiological history. The Ediacaran Period, in

fact, constitutes a dis-tinct chapter in thathistory, bounded belowby global ice ages andabove by the diversifi-cation of animal life—and characterized mostvividly by the unusual,mostly soft-bodied fos-sils that give it itsname. The unique mor-phologies of the Edia-cara biota have spawnedwidely varying system-atic interpretations—from giant protists andlichens to seaweedsand extinct experi-ments in multicellulari-ty. Most paleontolo-gists, however, agreethat the assemblage in-cludes early cnidarian-grade animals, as wellas burrows and trailsand perhaps body fos-sils of early bilateralorganisms (bilaterians)(3).

G E O L O G Y

A New Period for

the Geologic Time ScaleAndrew H. Knoll, Malcolm R. Walter, Guy M. Narbonne, Nicholas Christie-Blick

530

540Beginning of Cambriananimal diversification

Oldest calcified animals

Oldest animal burrows

Oldest macroscopicEdiacara biota

Gaskiers glaciation

Oldest animal embryos

Cap carbonate

Marinoan glaciation

Ag

e (M

a)

550

560

570

580

590

600

610

620

630

640

Ph

aner

ozo

icP

rote

rozo

ic

Cam

bri

anE

dia

cara

n“C

ryo

gen

ian

”

?

Looking for a few good rocks. (Left) Major events associated with

the Ediacaran Period. Brackets indicate uncertainty in the chronomet-

ric age of the GSSP. (Right) The formally defined base (GSSP) of the

Ediacaran Period in Enorama Creek, Australia, is located at the contact

of Marinoan glacial rocks and overlying Ediacaran cap carbonates (at

the right foot of the geologist).

A. H. Knoll is in the Department of Organismic andEvolutionary Biology, Harvard University, Cambridge,MA 02138, USA. M. R. Walter is at the AustralianCentre for Astrobiology, Department of Earth andPlanetary Sciences, Macquarie University, Sydney,NSW 2109, Australia. G. M. Narbonne is in theDepartment of Geological Sciences and GeologicalEngineering, Queen’s University, Kingston, OntarioK7L 3N6, Canada. N. Christie-Blick is in theDepartment of Earth and Environmental Sciencesand Lamont-Doherty Earth Observatory of ColumbiaUniversity, Palisades, NY 10964, USA. E-mail:[email protected]



622

Dates are important. The beginning of theperiod remains to be determined precisely,but the uranium-lead (U-Pb) zircon datingmethod gives a maximum age of 635.5 ± 1.2Ma for zircons from volcanic ash within gla-cial diamictites in Namibia (4). Meanwhile, aPb-Pb date of 599 ± 4 Ma for postglacialphosphorites from China (5) provides a min-imum age for the beginning of the EdiacaranPeriod. The earliest known animal fossils—microscopic eggs, embryos, and segmentedskeletal tubes—are found in the phosphorites(6). Following one last, regionally distributedglaciation, moderately diverse macroscopicfossils appear in ~575 Ma rocks fromNewfoundland (7). Bilaterian animal trailsenter the record no later than 555 Ma, andcalcified skeletons (of a distinctivelyEdiacaran, not Cambrian, aspect) by 549 Ma(8). Ediacaran assemblages persisted until theend of the period, separated from Cambriandiversification by a major, short-lived pertur-bation in the carbon isotopic record.

If Ediacaran fossils characterize the peri-od, why don’t they define it? The simple an-swer is that the fossils are scarce and, conse-quently, there are large uncertainties regard-ing correlation. Among sedimentary basins,the first appearance of Ediacara-type fossilscan differ by 10 million years or more. Thisis why the Ediacaran Period departs ratherabruptly from Phanerozoic convention indefining the beginning of the period by a cli-matic/geochemical event. The unusual de-pletion of 13C in the texturally striking car-bonates that veneer Marinoan glacial rocksis recognized globally and widely acceptedas a paleoceanographic signature of rapiddeglaciation, although mechanistic interpre-tations differ (9, 10). More generally, largesecular variations in the isotopic composi-tions of carbon, sulfur, and strontium havecome to play an important part in the corre-lation of Neoproterozoic (1000 to 543 Ma)sedimentary rocks. This works well becauseyounger Proterozoic strata record huge sec-

ular variations in the composition of seawa-ter that reflect not only global ice ages, butalso biospheric oxidation and global tecton-ic events. Indeed, the Neoproterozoic hasemerged as a primary focus of Earth sys-tems history, as scientists seek to understandthe complex interactions between planet andlife that gave rise to the Phanerozoic world.Testifying to this effort, the new EdiacaranPeriod provides a first extension of the geo-logic time scale into Earth’s Precambrianpast. It will not be the last.

References1. K. A. Plumb, Episodes 14, 139 (1991).

2. A. H. Knoll et al., Lethaia, in press.

3. G. M. Narbonne, GSA Today 8, 1 (1998).

4. K.-H. Hoffmann et al., Geology, in press.

5. G. H. Barfod et al., Earth Planet. Sci. Lett. 201, 203

(2002).

6. S. Xiao, A. H. Knoll, J. Paleontol. 74, 767 (2000).

7. G. M. Narbonne, J. G. Gehling, Geology 31, 27 (2003).

8. M. W. Martin et al., Science 288, 841 (1998).

9. P. F. Hoffman, D. P. Schrag, Terra Nova 14, 129 (2002).

10. G. Q. Jiang et al., Nature 426, 822 (2003).

In 1982 a team of U.S. scientists collectingmeteorites in Antarctica found a fragmentof the Moon. The 31-g meteorite, now

called Allan Hills (ALHA) 81005, had oncebeen a rock or a piece of a rock that existedat or near the Moon’s surface. At some timein the past, a meteoroid collided with theMoon and accelerated the rock to lunar es-cape velocity. After orbiting Earth for lessthan 200,000 years, the rock was captured byEarth’s gravitational field, landed inAntarctica, and was buried by snow. There itbecame a miniscule part of a huge glacier,which also carried other meteorites that hadfallen over the years. The glacier’s flow is im-peded by the Transantarctic Mountains, andnear the mountains meteorites are continual-ly exposed at the surface as wind and sun ab-late and sublime away the ice that encasesthem. The collecting team immediately rec-ognized that ALHA 81005 did not look likethe other meteorites that they were collecting,all of which were fragments of asteroids.Meteorite curators at the NASA JohnsonSpace Center, having seen a lot of Moonrocks from the Apollo missions, suspectedthat it was a Moon rock. Further studies haveconfirmed their suspicion (1). The stone wasthe first to be recognized as a lunar meteorite,

although three others not yet classified hadbeen collected in Antarctica 3 years earlier bya team from the Japanese National Instituteof Polar Research. Since 1979, about 30 lu-nar meteorites have been found, all in deserts.On page 657 of this issue, Gnos et al. (2) de-scribe the most unique lunar meteorite foundto date. This 206-g stone, known as Sayh alUhaymir (SaU) 169, was found in theSultanate of Oman in January 2002.

On the basis of the wide ranges in com-position, mineralogy, texture, and cosmic-ray exposure ages, the 30 lunar meteoriteslikely represent at least 20 impacts on thelunar surface, although the crater of originis not known for any of them. For any giv-en lunar meteorite, the fact that we don’tknow where on the Moon it originates is aserious detriment to geologic interpretationof data derived from the stone. However,the meteorites are samples from many ran-dom locations, and this characteristic pro-vides important information not availablefrom the Apollo samples, all of which werecollected on six missions to the centralnearside (see the figure).

P L A N E TA RY S C I E N C E

A Unique Chunk of the MoonRandy L. Korotev

The author is in the Department of Earth andPlanetary Sciences, Washington University in St.Louis, St. Louis, MO 63130, USA. E-mail: [email protected]

– 90°– 180° – 135° – 90° – 45° 0°

East longitude

No

rth

lati

tud

e

45° 90° 135° 180°

– 45°

– 0°

45°

90°

5 7

1624

– 0.3 0.9 2.1 3.3 4.5 5.7 6.9 8.1 9.3 10.5 11.7

Rich in radioactive elements. Distribution of thorium on the lunar surface [adapted from (5)].

Color scale shows thorium abundance in µg/g.The numbers represent the locations of the six Apollo

landing sites (1 = Apollo 11, 2 = Apollo 12, and so on; landing sites 2 and 4 are adjacent). The ellipse

indicates the position of the Imbrium basin. The center of the figure is the center of the nearside, as

viewed from Earth. Most of the Moon’s thorium and other incompatible elements are concentrat-

ed in the northwest quadrant of the nearside.



623

It came as a surprise to early lunar sam-ple researchers that samples from theApollo 12 and 14 missions had very highconcentrations of the suite of trace chemi-cal elements that geochemists categorize as“incompatible.” On the Moon, incompati-ble elements include phosphorus, the rareearth elements, and the three most impor-tant naturally occurring radioactive ele-ments, potassium, thorium, and uranium.Lunar rocks consist mainly of four miner-als. When those minerals crystallize from amagma, the incompatible elements are ex-cluded from the solid phases and are con-centrated in the liquid phase. The existenceof lunar rocks with high concentrations ofincompatible elements indicated that ig-neous differentiation of the Moon had oc-curred to an advanced degree and that theMoon, unlike the parent bodies of mostmeteorites, was not a primitive object.

Partial mapping of the lunar surface bygamma-ray spectrometers aboard Apolloorbiting command modules showed thatthe region in the vicinity of the Apollo 12and 14 sites was rich in radioactive ele-ments. However, it was not until 1998 thatthe first global geochemical coverage ofthe Moon was obtained by gamma-ray andneutron spectrometers aboard the LunarProspector mission (3–5). It only becameapparent more than 25 years after the lastApollo mission that three of the missions,Apollos 12, 14, and 15, had landed in a re-gion that was uniquely rich in incompatibleelements compared to most regions of theMoon and that the other three missionswere in near proximity to that geochemi-cally anomalous region (see the figure).

Various lines of evidence suggest thatthe early Moon was largely molten. TheLunar Prospector data showed that thechemical differentiation of the Moon, as itformed its core, mantle, and crust, wasasymmetric. The last liquid, by then rich inincompatible trace elements, concentratedin what is now the northwest quadrant ofthe nearside. One of the last and largestbasin-forming impactors, the one that pro-duced the Imbrium basin, struck this geo-chemically anomalous region 3.9 billionyears ago, spreading thorium-rich ejectaover the surface of the Moon (6). All sixApollo landing sites contain rocks, mainlyancient impact-melt breccias, that are richin thorium (typically 8 to 20 µg/g) and oth-er incompatible elements. One interpreta-tion is that these thorium-rich impact meltbreccias were produced when the Imbriumimpactor struck the incompatible-ele-ment–rich region (7). Another is that a cat-aclysmic set of impact events occurringwithin a short interval about 3.9 billionyears ago produced all of the major near-side lunar basins and that several of those

impacts excavated high-thorium material(8). Unlike typical rocks from the Apollosites, most lunar meteorites, including AL-HA 81005, have low concentrations of tho-rium, typically less than 1 µg/g. These low-thorium meteorites must originate from thevast portions of the Moon, mainly on thefarside, with low surface concentrations ofthorium (9).

SaU 169 represents the opposite ex-treme. It is an impact-melt breccia with ex-ceedingly high concentrations of thorium(33 µg/g) and other incompatible elements.As Gnos et al. argue, SaU 169 almost cer-tainly originates from within the high-thori-um anomaly. If lunar meteorites are randomsamples of the Moon, it was predictablethat sooner or later a high-thorium lunarmeteorite would be found. It is neverthelessironic that even though the Apollo missionsinadvertently visited some of the most tho-rium-rich areas of the Moon, SaU 169 isricher in incompatible elements than anyrock-sized Apollo sample.

The most significant aspect of the workof Gnos et al. is the 207Pb-206Pb crystalliza-tion age of 3.909 ± 0.009 billion years thatthey obtain for the impact melt on the basisof ion microprobe analysis of zircons. They

conclude that this age precisely dates theImbrium impact, although the age is signif-icantly older than the previous best work-ing ages of 3.85 ± 0.02 billion years (8) forImbrium and 3.89 ± 0.01 billion years forthe Serenitatis basin (10). The data for SaU169 call into question just which isotopicsystems best record the crystallization ageof an impact melt and whether the smalldifferences (<2%) in 40Ar-39Ar ages amongancient impact-melt rocks are significant.It is now imperative that 207Pb-206Pb agesbe obtained from zircons in thorium-richmelt breccias from the Apollo landing sitesfor comparison.

References1. U. B. Marvin, Geophys. Res. Lett. 10, 775 (1983).2. E. Gnos et al., Science 305, 657 (2004).3. D. J. Lawrence et al., Science 281, 1484 (1998).4. R. C. Elphic et al., J. Geophys. Res. 105, 20333

(2000).5. D. J. Lawrence et al., J. Geophys. Res. 108, 5102,

10.1029/2003JE002050 (2003).6. L. A. Haskin, J. Geophys. Res. 103, 1679 (1998).7. L. A. Haskin et al., Meteorit. Planet. Sci. 33, 959

(1998).8. G. Ryder, J. Geophys. Res. 107, 6-1, 10.1029/

2001JE001583 (2002).9. R. L. Korotev et al., Geochim. Cosmochim. Acta 67,

4895 (2003).10. G. B. Dalrymple, G. Ryder, J. Geophys. Res. 101, 26069

(1996).

Conventional engineering materialsare usually polycrystalline solidscomposed of crystallites (grains) that

are many micrometers in diameter (d). Therecent advent of nanocrystalline materials,for which d is less than 100 nm, has openednew opportunities for research and applica-tions. For the “upper nano” regime with d ≈100 nm, research on mechanical propertieshas focused on tailoring the grain and/orboundary structures for optimized strengthand ductility (1). The “lower nano” regimewith d < ~30 nm, on the other hand, is therealm for discovery of new deformationmechanisms (2–13). This is because the nu-cleation and movement of line defectscalled dislocations—the main carrier ofplastic deformation in coarse-grained met-als—is projected to be difficult in such tinygrains where the distance between disloca-tion pinning points becomes very small, de-manding very high stresses to activate dis-location sources. Deformation processesgoverned by abundant grain boundaries

may become important, as demonstrated innanocrystalline nickel by Shan et al. onpage 654 of this issue (2). Using an insitu dark-field transmission electron mi-croscopy (TEM) technique, the authorsrecorded the frequent rotation of nanocrys-tals (d ≈ 6 nm) into larger aggregates ofneighboring grains during deformation.

In general, a deforming grain is forcedto rotate in response to the external stressesexerted upon it by its neighbors. For con-ventional metals, grain rotation during de-formation often accompanies the formationof texture (preferred orientation) and is ac-complished by the microscopic dislocationglide on multiple active slip systems in thegrains. Such extensive dislocation activityis unlikely in 6-nm grains, and texturingwas not observed in heavily cold-rollednanocrystalline palladium (12). Grain rota-tion can also be caused by extensive grainboundary sliding and diffusion, which usu-ally occur only at elevated temperatures.But molecular dynamics (MD) computersimulations predict that such a mechanismwill become operative and even dominantwhen d is reduced to ~10 nm (4, 7, 8).Evidence for this has been sought for many

M AT E R I A L S S C I E N C E

Watching the Nanograins RollE. Ma

The author is in the Department of Materials Scienceand Engineering, Johns Hopkins University, Baltimore,MD 21218, USA. E-mail: [email protected]



624

years, but direct experimental confirmationremained elusive except in extremely thin(~20 nm) films used for high-resolutionTEM experiments (3).

The experimental findings of Shan et al.(2) are also valuable because our under-standing of the deformation mechanisms ofnanocrystalline metals has relied heavilyon MD simulations. With high loads andextreme strain rates applied to produce ameasurable strain within the subnanosec-ond MD time scale, the simulations mayexclude certain time-dependent processes

and cannot determine the true rate-limitingprocesses in real-world experiments. TheMD results are sources of inspiration andguidance but do not directly validate or dis-prove the existence of a mechanism (4, 9).

Taken together, recent TEM observa-tions (2, 4–6), in situ synchrotron x-ray dif-fraction (XRD) experiments (10), and ad-vances in MD simulations begin to paint aunifying picture for the unusual deforma-tion mechanisms in nanocrystalline face-centered-cubic metals (such as nickel andcopper) (7, 11). When dislocations are stillinvolved, their operation takes differentforms. First, the intragrain Frank-Read dis-location sources dominant in conventionalpolycrystals cease to control deformation.Instead, dislocations are nucleated out ofgrain boundary (GB) sources. There is alsoindirect evidence, including the increasedstrain rate sensitivity and unusually low ac-tivation volume, that hints at the defect(boundary)–assisted nucleation of disloca-tions as the thermally activated rate-control-ling process (14). For d on the order of 30nm, the dislocations traverse the grain and

disappear into the opposing GBs, with littlechance of storage inside the grains (4, 5).Second, partial dislocation emission from aGB becomes more favored at sufficientlysmall d. Indeed, estimates showed that theenergy cost for the nucleation of one partialdislocation at a time, which is observed inMD simulations for all the nanocrystallinemetals studied (including those with highstacking fault energy) (7, 9), can be lowerthan that for emitting a full (perfect) dislo-cation (6). The partial dislocation–mediatedprocesses do leave behind debris, such as

the stacking faults and defor-mation twins in the figure (6,12, 15), that can be observedin postmortem TEM. How-ever, the nucleation of a trail-ing partial dislocation on thesame plane of an existingleading partial dislocation,before the latter gets absorbedby the opposing GB, is oftenpossible and can be an ener-getically less expensiveprocess than other options[such as the emission of a sec-ond partial dislocation on theadjacent plane for twin nucle-ation (9)]. The trailing partialdislocation then erases thestacking fault and may catchup with the leading partialdislocation. The resulting fulldislocations are not often ob-served in TEM, as they leaveno footprint in their wake af-ter traversing the grain. Theiraction was inferred from the

in situ XRD measurements (10). It is inter-esting that a unit dislocation trapped insidethe grain under loading has been capturedby Shan et al. (2).

Large stresses are required to drive de-formation in nanoscale grains. The localstress intensity can be particularly high andvaried depending on loading conditions,such that one may observe all types of slip:extended partial dislocations forming stack-ing faults, full dislocations, and deforma-tion twinning (6, 9). Note that dislocationactivities persist even when d is reduceddown to ~10 nm (2, 6, 12, 15). At suchsmall grain sizes, GB sliding and grain ro-tation become detectable, concurrent withdislocation activities and possibly also ac-commodated by atomic shuffling at theGBs (4). A supporting finding for the activerole of grain rotation and GB sliding, as di-rectly observed by Shan et al., is the lack ofcrystallographic texture in nanocrystallinepalladium grains that remained equiaxedeven after large plastic deformation (12).Rosner et al. also suggest that the numberof glide systems active in the coplanar twin-

ning they observed is obviously short of thefive slip systems required for a general de-formation in polycrystals (12).

Simulations with idealized samples havepredicted predominant GB sliding at d <~10 nm (7), but the experimental samplesavailable so far [see figure 1 of Shan et al.(2)] always contain a grain size distributionand some impurities, with a large (numberand volume) fraction of the grains largerthan 10 nm. Some deposited grains incolumnar shape also may not be conduciveto GB sliding and grain rotation. As a result,GB–mediated plasticity may be a contribut-ing but not yet dominant mechanism. TheHall-Petch relationship (increasing strengthwith decreasing d) may continue to hold forthe nickel films of Shan et al. (13) withoutdisplaying an obvious maximum strengthbeyond which an inverse Hall-Petch behav-ior (softening) takes over (7).

Grain rotation and GB sliding, with noevidence of dislocation activities, werethought to control deformation in goldfilms with d = 10 nm (3), but the sampleused was ultrathin (10 to 20 nm) and thedeformation rate was very slow. In thiscase, the two-dimensional geometry lacksthe three-dimensional constraints and has avery high surface-to-volume ratio. This ac-centuates the thermally activated processesfacilitating grain rotation, such as diffu-sion, especially when under electron irradi-ation in TEM. Note that even the thickerfilms of Shan et al. are not free of such ef-fects (16), even though the authors believethat the surface effects and stress-assistedgrain growth (due to driving forces to re-duce surface or GB energy) are negligible(2). Therefore, although the result of Shanet al. brings new insight into the deforma-tion of extremely small grains, cautionshould be exercised before generalizing thebehavior in TEM foils as fully representa-tive of bulk deformation.

References and Notes1. Y. M. Wang et al., Nature 419, 912 (2002).

2. Z. Shan et al., Science 305, 654 (2004).

3. M. Ke et al., Nanostruct. Mater. 5, 689 (1995).

4. K. S. Kumar, H. Van Swygenhoven, S. Suresh, ActaMater. 51, 387 (2003).

5. R. C. Hugo et al., Acta Mater. 51, 1937 (2003).

6. M. W. Chen et al., Science 300, 1275 (2003).

7. J. Schiotz, K. W. Jacobsen, Science 301, 1357 (2003).

8. V. Yamakov et al., Nature Mater. 3, 43 (2004).

9. H. Van Swygenhoven et al., Nature Mater. 3, 399

(2004).

10. Z. Budrovic et al., Science 304, 273 (2004).

11. S. Cheng et al., Acta Mater. 51, 4505 (2003).

12. H. Rosner et al., Philos. Mag. Lett. 84, 321 (2004).

13. J. A. Knapp, D. M. Follstaedt, J. Mater. Res. 19, 218

(2004).

14. Our stress relaxation and jump tests for nickel with d

= 30 nm showed a room-temperature strain rate sen-

sitivity four times that of coarse-grained nickel and

an activation volume as small as 7 to 20 b3, where b

is the Burgers vector.

15. X. Z. Liao et al., Appl. Phys. Lett. 84, 592 (2004).

16. P. M. Derlet et al., Philos. Mag. A 82, 1 (2002).

2 nm

T

TT

T

TS S

Deformation debris. An example of features left by partial dis-

location–mediated processes during deformation [adapted

from (6)]. A Fourier-filtered high-resolution TEM shows the

stacking faults (S) and deformation twins (T) in a deformed

nanocrystalline aluminum grain (6). Note the mirror symmetry

across the twin boundaries, where deposited partial disloca-

tions are also observed.



The Pathophysiology of Mitochondrial Cell DeathDouglas R. Green1* and Guido Kroemer2*

In the mitochondrial pathway of apoptosis, caspase activation is closely linked tomitochondrial outer membrane permeabilization (MOMP). Numerous pro-apoptoticsignal-transducing molecules and pathological stimuli converge on mitochondria toinduce MOMP. The local regulation and execution of MOMP involve proteins from theBcl-2 family, mitochondrial lipids, proteins that regulate bioenergetic metabolite flux,and putative components of the permeability transition pore. MOMP is lethal because itresults in the release of caspase-activating molecules and caspase-independent deatheffectors, metabolic failure in the mitochondria, or both. Drugs designed to suppressexcessive MOMP may avoid pathological cell death, and the therapeutic induction ofMOMP may restore apoptosis in cancer cells in which it is disabled. The general rulesgoverning the pathophysiology of MOMP and controversial issues regarding its regula-tion are discussed.

T he major form of apoptosis seen inmost settings in vertebrate cells pro-ceeds through the mitochondrial path-

way, defined by a pivotal event in theprocess—mitochondrial outer membrane per-meabilization (MOMP). MOMP occurs sud-denly during apoptosis (1), leading to therelease of proteins normally found in thespace between the inner and outer mitochon-drial membranes (including cytochrome c,AIF, and others). Before, during, or afterMOMP, there is frequently a dissipation ofthe mitochondrial inner transmembrane po-tential (��m), and the timing of MOMPversus ��m loss can provide clues to themechanism involved in a particular setting.MOMP precipitates the death of the cellthrough as many as three general mecha-nisms, including the release of molecules in-volved in the activation of caspases that or-chestrate downstream events often associatedwith apoptosis, the release of molecules in-volved in caspase-independent cell death, andthe loss of mitochondrial functions essentialfor cell survival (table S1). A pathophysio-logical role for MOMP is emerging.

Mechanisms of MOMP and theDecision to DieThe mechanisms responsible for MOMPduring apoptosis remain controversial, al-though it is clear that many proteins caninhibit or prevent MOMP by local effects

on mitochondrial membranes (tables S2and S3). In general, two classes of mecha-nism have been described and each mayfunction under different circumstances:those in which the inner mitochondrialmembrane participates, and those involvingonly the outer membrane (Fig. 1).

In the first class of mechanism, a poreopens in the inner membrane, allowing waterand molecules up to �1.5 kD to pass through.Although most models of this pore, the per-meability transition (PT) pore, postulate rolesfor the adenine nucleotide transporter (ANT)in the inner membrane and the voltage-dependent anion channel (VDAC) in the out-er membrane (2), this is a hypothetical model(supporting online text). Recent evidence hasshown that the PT pore can form in theabsence of the ANT (3), and alternative mod-els accounting for this pore have been pro-posed (4). Opening of the PT pore can betriggered by multiple stimuli and leads to (i)��m loss as ions equilibrate across thismembrane, and (ii) swelling of the matrix aswater enters. The latter can result in sufficientswelling to break the outer membrane to pro-duce MOMP. It should be noted that althoughloss of ��m accompanies irreversible PT,many other events can produce this loss. Lossof ��m is not sufficient to prove the involve-ment of PT. Conversely, PT pore opening canbe transient (through flickering of the pore),and therefore sustained ��m does not pro-vide a firm argument against the involvementof the PT pore unless monitored continuouslythroughout MOMP. In view of the difficultiesof quantifying PT in living cells, and in theabsence of a clear molecular substrate for thepore (supporting online text), it may be apragmatic approach to define PT-associatedMOMP as a process that can be inhibited bysome ligands of putative PT pore constituents

such as VDAC, ANT, or the ANT-interactingprotein cyclophilin D (a target of cyclosporinA) (5). Ideally, methods that directly assessthe permeability of the inner membrane (6)should be employed.

The second class of mechanism forMOMP does not involve a major role for PTor the mitochondrial inner membrane. Thisappears to be mediated by members of theBcl-2 family of apoptosis-regulating proteinsacting directly on the outer mitochondrialmembrane (table S4). Anti-apoptotic Bcl-2family members function to block MOMP,whereas the various pro-apoptotic memberspromote it. The latter fall into two generalsubfamilies, based on sharing of Bcl-2 ho-mology (BH) domains. BH123 (or multido-main) proteins share BH1, BH2, and BH3and appear to be effectors of MOMP, becausecells from mice lacking the two major BH123proteins, Bax and Bak, fail to undergoMOMP in response to a wide range of apo-ptotic signals (7). The other subfamily, theBH3-only proteins (that contain only the BH3domains), can act either to activate Bax andBak or to interfere with the anti-apoptoticBcl-2 family members (8).

Vesicles composed of purified mitochon-drial outer membranes can be permeabilizedin response to activated forms of recombinantBax or Bid, the latter presumably actingthrough Bak on the outer membrane (9). Fur-ther, vesicles composed of mitochondrial lip-ids (without other mitochondrial proteins)were permeabilized by recombinant, mono-meric Bax, provided that active recombinantBid or a BH3 peptide derived from Bid waspresent. This generated openings of indeter-minate size that could not be visualized byconventional ultrastructural techniques. Suchopenings may be consistent with large lipidicpores composed of activated BH123 proteinsand lipids with potential for negative curva-ture in membranes (10), for instance, themitochondrial lipid cardiolipin. However, thepresence of cardiolipin in the mitochondrialouter membrane remains controversial.

Some studies have implicated the outermembrane protein VDAC in MOMP. Baxand Bak can bind to VDAC, but possiblywith different effects. Although the Bax-VDAC interaction is suggested to causeMOMP, interaction of Bak with VDAC-2appears to be inhibitory (11). One possibilityis that VDAC functions to sequester smallamounts of cardiolipin or related lipids

1Division of Cellular Immunology, La Jolla Institute forAllergy and Immunology, 10355 Science Center Drive,San Diego, CA 92121, USA. 2Centre National de laRecherche Scientifique, Unite Mixte de Recherche8125, Institut Gustave Roussy, 39 rue Camille-Desmoulins, F-94805 Villejuif, France.

*To whom correspondence should be addressed. E-mail: [email protected] (D.R.G.) and [email protected] (G.K.)

REVIEW


present in the outer membrane to microdo-mains in which local concentrations of theselipids may be sufficient to allow permeabili-zation of the membrane by activated Bax orBak. This would also account for the apparentbinding of Bax to VDAC.

PT-independent MOMP can be followed bysecondary PT. In sympathetic neurons deprivedof nerve growth factor, Bax-dependent, PT-independent MOMP associated with cyto-chrome c release causes caspase activation andapoptosis. However, in the presence of caspaseinhibitors, such cells survive until ��m drops(12). Studies in which cyclosporin A blocks the��m loss and commitment to death suggestthat PT determines the point of no return inthese cells (13).

Upon MOMP, proteins of the intermem-brane space are released, although whether ornot all proteins are released simultaneouslyremains controversial. One suggestion is thatthese proteins are differentially sequesteredin the intermembrane space and that second-ary events are required for the release ofsome of them (14, 15). For example, remod-eling of the matrix and inner mitochondrialmembrane may be required for the release ofcytochrome c in some cases (15), although inother cases this was not observed (16). Suchremodeling has been suggested to be mediat-ed by PT (15). Further, mitochondrial fissioncan occur around the time of MOMP (17)and proteins that regulate fusion or fission ofmitochondria appear to affect which proteinscan be released upon MOMP.

Irrespective of its mechanisms, MOMPcan seal the point of no return of the lethalprocess by the release of caspase activatorssuch as cytochrome c (table S1). Once acti-vated, caspases can cause a rapid loss ofmitochondrial functions. Upon MOMP, exe-cutioner caspases can cleave the NDUSF1subunit of respiratory complex I, and muta-tion of its single cleavage site can preservemitochondrial functions during apoptosis(18). This can delay plasma membrane eventsassociated with caspase activation, includingloss of plasma membrane integrity and exter-nalization of phosphatidylserine, thus indicat-ing an important role for disruption of mito-chondrial function in apoptotic cell death.Nonetheless, even without caspase activation,MOMP generally results in cell death throughthe release of multiple caspase-independentdeath effectors, as well as loss of essentialmitochondrial functions (table S1 and sup-porting online text).

Upstream of MOMPMultiple distinct signaling pathways con-verge on MOMP (tables S2 and S3). Al-though some of the BH3-only proteins inthe Bcl-2 family have the capacity to acti-vate Bax and Bak or, conversely, inhibit theanti-apoptotic Bcl-2 family members, other

molecules may have these properties aswell. The tumor suppressor p53 acts, inpart, to induce apoptosis by inducing ex-pression of the BH3-only protein PUMA,and PUMA-deficient cells display a resis-tance to p53-mediated apoptosis (19, 20).However, p53 can trigger MOMP and ap-optosis in the absence of transcription, andthis can occur through direct activation ofBax (21) or Bak (22) or through binding toBcl-2 and Bcl-XL, which blocks their ac-tivity (21, 23). Resolving the role for thismechanism versus that of transcriptionalregulation will be important in understand-ing the apoptotic function of p53.

An emerging theme is one of nuclearproteins functioning in the cytosol throughdirect interactions with Bcl-2 family pro-

teins. Ku70, involved in DNA repair, caninhibit Bax (24 ). Another nuclear protein,TR3, binds Bcl-2 and perhaps promotesMOMP through this interaction (25). His-tone 1.2, released from the nucleus uponX-ray–induced DNA damage, can triggerMOMP (26 ), perhaps through an interac-tion with Bcl-2 family members. Hexoki-nase can interact with VDAC, and thisinteraction may inhibit the ability of Bax tocause MOMP (27 ). Intriguingly, enforcedexpression of hexokinase together with theglucose transporter Glut-1 is sufficient toconfer cell survival (28).

Alternatively, Bcl-2 family members mayact independently of mitochondria and up-stream of MOMP. Cells lacking Bax and Bakdisplay reduced calcium efflux from the en-

BH3-only and other proteins

BH-123 proteinsAPOPTOSIS-INDUCINGSIGNALS

No involvement ofinner membrane

APOPTOSIS-INDUCINGSIGNALS

Permeability transition

Bax

Bak

VDAC

ANT

OMM

IMMIMS proteinsreleased

Matrix swells,breaks OMM

water enters,∆Ψm dissipates

Pore opens

OMM

IMM

Bak

Bax

IMM

OMM

Bak

Bax

BH3-only and otherproteins

Bcl-2

activator

derepressor

A

BC

OMM

Fig. 1. Mechanisms for MOMP during apoptosis. (A) Signals for the induction of apoptosis (top)engage the activities of a subgroup of pro-apoptotic, BH3-only members of the Bcl-2 protein familyand other proteins, which in turn activate the pro-apoptotic, BH123 proteins Bax and Bak tooligomerize and insert into the outer mitochondrial membrane (OMM). Other BH3-only proteinscan act indirectly by releasing the first subgroup of BH123-activators from the anti-apoptotic Bcl-2family proteins that sequester them. The BH123 proteins engage either of the two mechanismsthat follow, perhaps depending on cell type or other conditions. (B) In PT-dependent MOMP,apoptosis-inducing signals act directly or indirectly to open the putative PT pore. This is composedof ANT or other proteins in the inner mitochondrial membrane (IMM) and is associated with VDACand perhaps other proteins in the OMM. Opening the pore allows water to enter the matrix andions to equilibrate, dissipating ��m at least transiently. The matrix swells, rupturing the OMM torelease proteins of the mitochondrial intermembrane space (IMS). (C) In PT-independent MOMP,BH123 proteins, perhaps with other proteins, cause the formation of pores in the OMM throughwhich IMS proteins are released. The mitochondrial tomograph was a gift from G. Perkins and D.Newmeyer. This was modified in the illustration.

R E V I E W


doplasmic reticulum in response to somestimuli (29) [whether this is directly an effectof Bax and Bak or an indirect effect of Bcl-2on the receptor is controversial (30)]. Theresulting calcium flux may act on mitochon-dria to produce MOMP independently of Baxand Bak, by induction of PT (29).

The Mitochondrial Pathway ofApoptosis in PathogenesisMOMP-dependent apoptosis is involved inmajor pathologies, with far-reaching medicaland pharmaceutical implications (Table 1 andtable S5). However, the role of MOMP indisease is often inferred from correlativestudies, such as when a disease-associatedmolecule or pharmaceutical agent is shown tohave effects on mitochondria, and shouldtherefore be treated with caution. Neverthe-less, the manifestations of MOMP (the mito-chondrial release of intermembrane proteins,as well as the dissipation of ��m) are fre-quently observed in disease states with in-creased cell death (31).

Many viruses have acquired the capacity tointercept or to activate the principal signal-transducing pathways leading to cell death.Several proteins from pathogenic viruses aretargeted to mitochondria and induce MOMP(Table 1 and table S5), and at least one is avirulence factor; a mutation in the human im-munodeficiency virus (HIV) Vpr protein thatattenuates its MOMP-inducing activity is statis-tically associated with a reduced risk to developacquired immunodeficiency syndrome (AIDS)(31). Several oncogenic viruses encode MOMP-inhibitory proteins, and in humans these may beinvolved in the formation of virally inducedlymphomas or Kaposi’s sarcoma.

In acute pathologies, for instance afterischemia, a combination of increased intra-cellular Ca2�, reactive oxygen species, and

metabolic perturbations can trigger MOMP,which ultimately accounts for cell loss incardiac infarction and cerebral stroke. Thiscell loss involves acute necrosis in the isch-emic core and a slower apoptosis in thepenumbra. The hypocampal CA1 region,extremely vulnerable to ischemia, containsmitochondria with the highest susceptibili-ty to Ca2�-induced MOMP in vitro (32).Key modulators of apoptosis, such as p53and Bax, facilitate ischemia-inducedMOMP and neuronal death (33), whereasthe MOMP inhibitor Bcl-2 can preventischemia-induced neuronal apoptosis.

MOMP is also likely to be involved inchronic neurodegenerative diseases. For ex-ample, in Huntington’s disease, polyglu-tamine expansions in huntingtin trigger neu-ronal death, and this aberration correlateswith huntingtin-induced mitochondrial ab-normalities. Transgenic overexpression ofBcl-2 can prolong the life span of mice car-rying an SOD-1 mutation found in patientswith familial amyotrophic lateral sclerosis, adegenerative disease affecting spinal motorneurons. Stabilization of mitochondrial mem-branes by genetic or pharmacologic manipu-lations also suggests a role for MOMP inneurodegeneration (table S5).

MOMP is also involved in acute toxin-induced cell death. Toxins implicated in Reye’ssyndrome, including salicylate, adipic, isovaler-ic, 3-mercaptoproprionic, 4-penenoic, and val-proic acids, cause MOMP when added to puri-fied mitochondria or to hepatocytes. Otherprominent examples of toxic MOMP inducersinclude ethanol, CCl4, and heavy metals or theirorganic derivatives (table S5).

MOMP induction is a therapeutic goal incancer therapy. MOMP is regulated by sev-eral oncogene products, in particular the anti-apoptotic Bcl-2 family proteins, whereas sev-

eral tumor suppressors induce or favorMOMP. A relative resistance to MOMP is aprominent hallmark of cancer (34). Experi-mental drugs that act on mitochondrial pro-teins or lipids have been shown to be thera-peutic in preclinical mouse models (table S5),and some therapeutic treatments have beenreported to induce MOMP, although it is notclear if direct MOMP induction accounts fortheir anticancer effects (table S6).

There is a great interest in developingdrugs that prevent MOMP and suppresspathological cell death (table S5). Undersome circumstances, MOMP is delayed orinhibited by cyclosporin A (CsA), and thiscan reduce the lethal effects of heavy met-als or high-dose paracetamol in animalmodels (35). CsA or Bcl-2 can reduce in-farct size in the heart and brain (2). CsAcan also be used to enhance the functionalrecovery after hypothermic heart preserva-tion (36 ). Several neuroprotective drugsalso prevent Bax-mediated MOMP in iso-lated mitochondria: tauroursodeoxycholicacid probably through inhibition of Baxinsertion (37 ) and dibucaine and proprano-lol at a later step that may involve outermembrane lipids (38).

Concluding RemarksOne particularly intriguing aspect thatemerges from the complexity of MOMPregulation is the functional and/or physicalinteraction between apoptosis regulators(e.g., the Bcl-2 family) and proteins knownto participate in intermediate metabolism,e.g., VDAC, hexokinase, or glucokinase.For example, the latter interacts with thepro-apoptotic protein Bad (39). These in-teractions may tie specific metabolic de-mands to apoptotic control and thus deter-mine “metabolic windows” for cells to

Table 1. Examples of pathogenic processes involving excessive or deficient MOMP.

Disease Pathogenic perturbation of MOMP Pharmacological correction of deregulated MOMP

Ischemia reperfusion damageof brain or heart

Redox stress, excessive Ca2� load, absent adenosinetriphosphate and nicotine adenine denucleotide,and accumulating fatty acids favor PT and MOMP.

Bcl-2 inhibitors of the PT pore, as well as mito KATP channelopeners, can exert neuro- or cardioprotective effects.

Neurodegenerative diseases Respiratory dysfunction affects highly sensitiveneurons in the central nervous system, leadingto their premature death.

Putative inhibitors of the PT pore (minocyclin, rasagiline, andtauroursodeoxycholic acid) can prevent neurodegeneration.

Liver disease Hepatotoxins (including bile acid and ethanol)and hepatitis B or C–encoded proteins induceMOMP.

Ursodeoxycholic acid prevents bile acid–induced PT andthus exerts hepatoprotective effects.

Cancer MOMP-inhibitory proteins from the Bcl-2family or unrelated proteins (such as Muc1)enhance apoptosis resistance.

Cytotoxic agents targeting Bcl-2–like proteins, PT porecomponents, and/or mitochondrial lipids enforce MOMPand kill cancer cells.

HIV-1 infection Vpr, an accessory HIV-1–encoded protein, canact on ANT and Bax to trigger MOMP. Thiseffect is frequently lost because of a mutationin long-term nonprogressors.

HIV-1 protease inhibitors can inhibit MOMP induced byVpr in isolated mitochondria.

R E V I E W


avoid MOMP. In the absence of growthfactors that regulate metabolite flow (40) orin conditions distant from optimal metabol-ic conditions (oxygen tension, redox poten-tial, tissue pH, and glycolytic substrates),cells may be primed for MOMP and de-mise. This crosstalk between apoptosis andmetabolism may contribute to the metabol-ic signature of cancer, the Warburg phe-nomenon, an increased reliance on anaero-bic metabolism even in the presence ofabundant oxygen. Progress at the frontiersof pathobiology will help to integrate theprocess of MOMP and its regulation intothe physiology of the cell.

References and Notes1. J. C. Goldstein, N. J. Waterhouse, P. Juin, G. I. Evan,D. R. Green, Nature Cell Biol. 2, 156 (2000).

2. M. P. Mattson, G. Kroemer, Trends Mol. Med. 9, 196(2003).

3. J. E. Kokoszka et al., Nature 427, 461 (2004).4. L. He, J. J. Lemasters, FEBS Lett. 512, 1 (2002).5. P. C. Waldmeier, K. Zimmermann, T. Qian, M. Tintelnot-Blomley, J. J. Lemasters, Curr. Med. Chem. 10, 1485(2003).

6. D. Poncet, P. Boya, D. Metivier, N. Zamzami, G. Kro-emer, Apoptosis 8, 521 (2003).

7. M. C. Wei et al., Science 292, 727 (2001).8. A. Letai et al., Cancer Cell 2, 183 (2002).9. T. Kuwana et al., Cell 111, 331 (2002).10. G. Basanez et al., J. Biol. Chem. 277, 49360 (2002).11. E. H. Cheng, T. V. Sheiko, J. K. Fisher, W. J. Craigen,

S. J. Korsmeyer, Science 301, 513 (2003).12. M. Deshmukh, K. Kuida, E. M. Johnson Jr., J. Cell Biol.

150, 131 (2000).13. L. K. Chang, R. E. Schmidt, E. M. J. Johnson, J. Cell Biol.

162, 245 (2003).14. M. Ott, J. D. Robertson, V. Gogvadze, B. Zhivotovsky,

S. Orrenius, Proc. Natl. Acad. Sci. U.S.A. 99, 1259(2002).

15. L. Scorrano et al., Dev. Cell 2, 55 (2002).16. O. von Ahsen et al., J. Cell Biol. 150, 1027 (2000).17. M. Karbowski, R. J. Youle, Cell Death Differ. 10, 870

(2003).18. J. E. Ricci et al., Cell 117, 773 (2004).19. J. R. Jeffers et al., Cancer Cell 4, 321 (2003).20. A. Villunger et al., Science 302, 1036 (2003).21. J. E. Chipuk et al., Science 303, 1010 (2004).22. J. I.-J. Leu, P. Dumont, M. Hafey, M. E. Murphy, D. L.

George, Nature Cell Biol. 6, 443 (2004).23. M. Mihara et al., Mol. Cell 11, 577 (2003).24. M. Sawada et al., Nature Cell Biol. 5, 320 (2003).25. B. Lin et al., Cell 116, 527 (2004).26. A. Konishi et al., Cell 114, 673 (2003).27. N. Majewski, V. Nogueira, R. B. Robey, N. Hay, Mol.

Cell. Biol. 24, 730 (2004).

28. J. C. Rathmell et al., Mol. Cell. Biol. 23, 7315 (2003).29. L. Scorrano et al., Science 300, 135 (2003).30. M. J. Thomenius, C. W. Distelhorst, J. Cell Sci. 116,

4493 (2003).31. A. D. Badley, G. Kroemer, Trends Pharmacol. Sci. 24,

298 (2003).32. G. Mattiasson, H. Friberg, M. Hansson, E. Elmer, T.

Wieloch, J. Neurochem. 87, 532 (2003).33. G. V. Putcha et al., Neuron 38, 899 (2003).34. D. R. Green, G. I. Evan, Cancer Cell 1, 19 (2002).35. D. Haouzi et al., Apoptosis 7, 395 (2002).36. K. G. Rajesh, S. Sasaguri, S. Ryoko, H. Maeda, Trans-

plantation 76, 1314 (2003).37. C. M. Rodrigues et al., Proc. Natl. Acad. Sci. U.S.A.

100, 6087 (2003).38. B. M. Polster, G. Basanez, M. Young, M. Suzuki, G.

Fiskum, J. Neurosci. 23, 2735 (2003).39. N. N. Danial et al., Nature 424, 952 (2003).40. J. C. Rathmell et al., Mol. Cell. Biol. 23, 7315

(2003).41. Supported by Agence Nationale pour le Recherche

sur le SIDA, European Commission, Ligue Nationalecontre le Cancer, and the French Ministry of Science(G.K.) and by NIH and Gemini Science (D.R.G.).

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/626/DC1SOM TextTables S1 to S6References and Notes

R E V I E W


631www.sciencemag.org SCIENCE VOL 305 30 JULY 2004

T E S T I N G H U M A N L I M I T S SP

EC

IAL

SE

CT

ION

In the 2 millennia since the first Olympic games, one principle has withstood the test of time:People are obsessed with pushing the human body to the limit. With the curtain set to rise nextmonth in Athens on the latest Olympic Games, this Special Section goes backstage to exploresome of the defining attributes of the world’s greatest athletes—and their Achilles’ heels.

Doping allegations against elite athletes have cast a long shadow in the run-up to this year’sgames. In a News report (p. 632), Vogel examines how scientific sleuths are devising new

methods to unmask athletes bent on cheating their way to the top. An STKE Perspective byHandelsman (www.sciencemag.org/sciext/sports) assesses the poorly

defined physiological roles of designer androgens such as THG. And in aSAGE KE News Synthesis article (www.sciencemag.org/sciext/sports),

Davenport probes the promises and pitfalls of growth hormone. What does it take to swim the fastest, throw a discus the far-

thest, or jump the highest? In some sports, it would seem, ath-letes claim such honors by birthright. Men and women fromKenya’s Rift Valley dominate endurance running, for example,and runners from West Africa reign supreme as sprinters.

Holden explores these apparent genetic edges in a News report(p. 637). The presence or absence of a Y chromosome creates adifferent kind of uneven playing field. A decade ago the bestfemale runners were closing in on the times of their malecounterparts. But Holden reports (p. 639) that the gender gap

has plateaued or even increased over the past 15 years in allrunning events apart from the marathon.

Tending not to discriminate by gender are injuries sus-tained from pushing the limit. In a News report (p. 641),

Stokstad describes how young gymnasts maybe raising their risk of osteoarthritis and otherhealth problems later in life. On a more posi-

tive note, information gained from studyinghow athletes’ muscles respond to training is pro-

viding new insights on muscle growth and atrophy,the topic of an STKE Review by Sartorelli (www.

sciencemag.org/sciext/sports). Ultimately, performance comes down to mechanics.

On page 643, Cho profiles Mont Hubbard, a mechani-cal engineer who has spent his career optimizing motion in

sports. New materials can reduce physical constraints to performance. At nextmonth’s games, many of the world’s best swimmers will be wearing suits withtiny ridges modeled on sharkskin that are claimed to reduce friction and drag.In a News report (p. 636), Krieger investigates those claims.

Sports scientists may not share the limelight with their study subjects, butthe discipline is gathering momentum, as overview articles and profiles of early-career researchers attest on Science’s Next Wave (www.sciencemag.org/sciext/sports). These experts toil behind the scenes of a pursuit that, in oneform or another, captivates most everyone’s attention—especially every 4 years.

–RICHARD STONE

From the Ignoble to the Sublime

T I T L E O F S P E C I A L S E C T I O N

I N T R O D U C T I O N

CR

ED

IT:N

AY

OM

I K

EV

ITIY

AG

ALA

/SC

IEN

CE

C O N T E N T S

N E W S

632 A Race to the Starting LineMighty Mice: Inspiration for Rogue Athletes?

636 Do Pool Sharks Swim Faster?

637 Peering Under the Hood ofAfrica’s Runners

639 An Everlasting Gender Gap?

641 Graceful, Beautiful, and Perilous

643 Engineering Peak PerformanceLong Gone or Gone Wrong?

See related STKE, SAGE KE, and NextWave material on p. 567 and Editorial onp. 573.


KREISCHA, GERMANY—Tucked on the woodededge of this village in the Saxon hills southof Dresden is a drab, single-story officebuilding with a sinister past. Until 1989 of-ficials of the German Democratic Republictested their athletes here to certify them asdrug-free before international competitions.But it was all a charade. Many of the EastGerman athletes, both men and women,were systematically doped up with testos-terone and other anabolic steroids, oftenwithout their knowledge. It was the Kreis-cha lab’s responsibility to ensure that theregimen was suspended long enoughbefore a competition to flush out anytraces of drugs, explains Klaus Müller,director of the Institute for DopingAnalysis and Sport Biochemistry thattoday occupies the building. Some-times the drug docs cut it too close.“You would hear that a certain famousathlete couldn’t travel to a competitionbecause of a ‘sudden illness,’ ” saysMüller, whose institute is part of aworldwide antidoping network. “Weall knew what that meant.”

That crooked chapter in Germansport is over, but the practice of dop-ing appears to be more widespreadthan ever. Last month world championsprinter Kelli White received a 2-yearban from competition after admittingto having taken banned steroids andthe hormone erythropoietin (EPO),which boosts red blood cell counts.Other clients of the Bay Area Labora-tory Co-operative (BALCO) nutritioncenter in Burlingame, California, were also implicated in the scandal; asScience went to press, U.S. officialswere investigating evidence thatOlympic gold medalists Marion Jonesand Chryste Gaines and world-recordsprinter Tim Montgomery had beentreated with banned steroids and hor-mones by the same lab.

In the privileged world of elitesports, avarice and the pursuit of glory con-tinue to lead coaches and chemists astrayand tempt athletes to risk health and medals.“Sport can be so magnificent and so power-ful precisely because humans play the keyrole,” says Andrew Pipe, a physician at theUniversity of Ottawa Heart Institute and for-mer chief medical off icer to Canada’sOlympic team. “It can be so depressing and

sordid for exactly the same reason.” Dopers are getting better at covering

their tracks, forcing researchers to inventnew techniques to detect ever more subtleuses of synthetic chemicals or proteins thatboost the body’s ability to build muscle,shed fat, or carry oxygen. What was oncethe exclusive domain of analyticalchemists—who searched for steroids inurine samples—now involves endocrinolo-gists and geneticists as authorities attempt toclamp down on what could become the nextillicit frontier: doping with genes for muscle

building. “Testing gets better and better, butthe opposition gets better and better too,”says Don Catlin, director of the Universityof California, Los Angeles (UCLA),Olympic Analytical Laboratory.

Back-alley chemistryAthletes have been seeking an artificialedge since at least the late 1800s, when run-

ners and long-distance bicyclists used nitro-glycerin and even cocaine to boost staminaand block pain. But although authorities be-gan testing for banned substances in the1970s, their efforts had little impact, saysPeter Sonksen of St. Thomas’ Hospital inLondon and a former member of the Inter-national Olympic Committee’s (IOC’s) med-ical commission. “For a long time there wasa feeling that many sporting bodies wereprotecting their players,” he says.

A lack of vigilance created an environ-ment for blatant cheating. For example, aseries of astounding world-record per-formances in the 1980s, especially in power sports such as the shot put or ham-mer throw, were almost certainly fueled bytestosterone and other prohibited anabolicsteroids, Müller says.

There is little doubt that steroids help ath-letes beef up. By targeting the samereceptor as testosterone does, theyboost the body’s capacity for buildingmuscle and erode its capacity forbreaking it down. But they have man-ifold side effects. Although womenproduce some testosterone naturally,ratcheting up levels even slightlyleads to increased body hair and acneand can wreak havoc with the repro-ductive system. In men, takingsteroids suppresses natural produc-tion of testosterone, which can lead tobigger breasts, shrunken testicles, andinfertility. In both sexes, high dosesof the drugs damage the liver and thecardiovascular system.

As testing for steroids began tobe enforced more strictly in the1990s, use of the drugs plummeted—and the pace of record-breakingtapered off. The antidoping forcesseemed to have the upper hand until2002, when the sport world wasrocked by revelations that a pair ofso-called designer steroids—drugswith no legitimate medical use—had been synthesized, apparently toelude doping testers.

In one case Catlin’s team detectedunusually low levels of naturalsteroids such as testosterone,epitestosterone, and androsterone inthe urine of a female cyclist, a sign

that something was amiss. Probing further,his group found traces of norbolethone, anandrogen developed by Wyeth in the 1960s.In animal tests, Catlin says, norbolethoneappeared to be a very effective musclebuilder while having relatively few mas-culinizing side effects. It was tested in shortchildren and underweight patients, butWyeth shelved the compound, apparently C

RED

IT:S

OV

FOTO

/EA

ST

FOTO

T E S T I N G H U M A N L I M I T S

SP

EC

IAL

SE

CT

ION

N E W S

A Race to the Starting LineScientists are scrambling to devise new methods for snaring athletes whocheat with steroids, hormones, and, someday, even extra genes

Image not available for online use.

Tainted glory. World champion Marita Koch was never directly

implicated, but many of her East German teammates were given

performance-enhancing drugs by their doctors and coaches.

because of toxic side effects. Evidently,someone was cooking up a new supply.

A whistleblower made the second dis-covery possible. In June 2003 Catlin received the residue from a used but emptysyringe from the U.S. Anti-Doping Agency.A track coach had sent it to the authorities,suggesting that they take a careful look.Within a few weeks, Catlin and his col-leagues had identified tetrahydrogestrinone(THG). The new chemical, which had neverbefore been described, resembles twosteroids banned for use by professional ath-letes: gestrinone, prescribed occasionallyfor the treatment of endometriosis, andtrenbolone, which has some uses in veteri-nary medicine. Both steroids have powerfulanabolic effects, and the UCLA teamquickly suspected that the derivative hadbeen designed to activate the same recep-tors while foiling standard screens forknown steroids. When authorities testedurine stored from previous competitions,they found at least a dozen THG-taintedsamples, many from athletes who had con-nections to BALCO.

Because routine screening would never

have caught THG, doping testers were con-fronted with the prospect of having to de-velop ways to detect an incalculable arrayof steroids and other chemicals that mightplay a similar performance-enhancing role.“The THG story tells us very convincinglythat there are people out there who arescheming to develop new entities to give toathletes,” says Catlin. “We’ve studied thechemistry, and there’s essentially no end tothe possibilities. Are there others out there?There certainly are.” They just haven’t beenidentified yet, he adds.

Some labs are hoping to defeat dopers attheir own game. Wilhelm Schänzer and hiscolleagues at the Institute of Biochemistryof the German Sport University Colognehave begun churning out more than a dozenpotential designer agents by tinkering withexisting steroids. “We’re trying to think inthe same way as those who are trying tomake new compounds,” Schänzer says. Hisgroup uses mass spectrometry to profile theconcoctions and identify signals that mightbetray illicit compounds in bodily fluids.

THG presented a legal challenge aswell. Lawyers for athletes who tested posi-

tive argued that the authorities couldn’tdemonstrate that the substance is an ana-bolic steroid, and therefore it could not beclassified as a banned substance. Indeed,the chemical’s effects in animals—muchless humans—had never been character-ized in a legitimate lab; standard animaltests take many months. Under court-imposed time constraints, scientists resort-ed to a quicker solution, a test originallydesigned to ferret out environmental pollu-tants that mimic hormones. The test usesyeast cells altered to make the human ver-sion of the testosterone receptor as well asa luminescent protein that glows when thereceptor is activated. Using the test, DavidHandelsman of the ANZAC Research In-stitute in Concord, Australia, found thatTHG lights up the cells more brightly thanstandard anabolic steroids such as tren-bolone and even testosterone.

The confirmation came just in time tosupport the case against European champi-on sprinter Dwain Chambers, who had test-ed positive for THG in August 2003.(Chambers has said that he ingested thecompound unknowingly in a supplement


CR

ED

IT:N

.RO

SEN

TH

AL/

EU

RO

PEA

N M

OLE

CU

LAR

BIO

LOG

Y L

AB

OR

ATO

RY

,ITA

LYT E S T I N G H U M A N L I M I T S S

PE

CIA

LS

EC

TIO

N

Mighty Mice: Inspiration for Rogue Athletes?The mice seem indestructible. First described in 2001, the

Schwarzenegger mice, as they were dubbed by the press, have

twice as much muscle as normal mice, live longer, and can recover

from injuries that kill their weaker cousins. They build muscle with-

out exercising, and they seem to defy the aging process. “As they

age, they don’t get weaker,” says Nadia Rosenthal, a developmental

geneticist at the European Molecular Biology Laboratory in

Monterotondo, Italy, who created the mighty mice.

Rosenthal and her colleagues are hoping that the animals, which

carry an extra copy of a gene that codes for a protein called insulin-

like growth factor-1 (IGF-1), will help them understand and treat

muscle-wasting diseases such as muscular dystrophy. The mice may

also provide insights into wound healing and the mysterious process

of regeneration. But sports authorities are worried: Can the technique

that makes supermice be used in hu-

mans to create superathletes?

The practice may have already

begun. Although there is no test for

the molecule, a ready supply is out

there. Genentech manufactured

and tested IGF-1 in the 1980s but

decided not to market it. Tercica, a

company in South San Francisco,

California, is sponsoring clinical tri-

als of growth hormone plus IGF-1

in short children who don’t respond

to treatment with growth hormone

alone. “If someone is doing it legal-

ly, you can bet they’re doing it ille-

gally,” Rosenthal says.

But there’s a subtlety to the

protein that should deter any athlete tempted to inject the bottled

version. Some IGF-boosted mice, rather than being mighty, are in

fact rather sickly. It seems that the body makes at least four forms

of IGF-1. One circulates in the bloodstream and suppresses the

production of human growth hormone. Another is produced in

muscle tissue and appears in response to injury. Scientists are still

sorting out the differences among the forms, but it’s the second

type that Rosenthal, H. Lee Sweeney of the University of Pennsyl-

vania in Philadelphia, and their colleagues boosted in their mighty

mice. When production of the first form is increased, mice develop

oversized and weak hearts and are prone to cancer, Rosenthal says.

The form that Tercica is testing and athletes might be using is a

stripped-down version of the protein, without any of the distinguish-

ing alterations that the naturally made forms carry. It isn’t yet clear

how its effects compare with those of the naturally made versions,

but Rosenthal says it still might be tempting to athletes: Preliminary

data suggest that the Tercica version does encourage muscle growth.

What worries doping testers

most, however, is the possibility

that the gene-therapy technique

that created the mighty mice might

eventually be attempted in athletes.

Such a treatment would be difficult

to unmask, because the doped-up

IGF-1 gene, designed to remain in

muscle cells where it is produced,

would not be detectable in blood or

urine, says Rosenthal. Athletes

would understandably be reluctant

to give muscle biopsies just before a

competition. That leaves yet anoth-

er conundrum for testers to resolve

in the fight against doping.

–G.V.

Wanna race? A mouse with an extra copy of IGF-1 (right)

dwarfs its wild-type counterpart.


provided by BALCO.) In February, U.K.Athletics banned him from running in com-petitions for 2 years. Chambers had beenconsidered a favorite for the gold medal thissummer in Athens, but according to BritishOlympic Association rules, he is bannedfrom the Athens games.

The bioassays may soon join a growingarsenal that scientists are assembling tothwart the use of new designer steroids,says Handelsman. He and his col-leagues, for example, are working on asimple test to compare the amount oftestosterone normally present in theurine of men and women with the totalsteroid load, as measured by the bio-assays. “If there’s a gap, then that sug-gests there’s an unidentified substancethere,” Handelsman says.

The workhorse of steroid detection,the mass spectrometer, could also be putto innovative use. Even if an analysis failsto flag unexpected side chains or telltalepeaks, it can reveal subtle differences inthe ratio of carbon isotopes that can helpidentify the origin of organic molecules.An unusual ratio of carbon-12 to carbon-13 in certain molecules can raise a redflag in a doping test. If a steroid moleculehas a ratio typical of a plant rather than ananimal, it is a sign that it comes from anoutside source, says Schänzer.

In pursuit of oxygenUnknown steroids are hard enough topin down; injections of naturally occur-ring hormones are even more elusive.Hormone levels fluctuate from hour tohour and from person to person, someasuring absolute amounts can’t naila doper. To do that, scientists must findsecondary signals indicating that thebody’s normal chemistry has been tam-pered with.

For years, some athletes took advan-tage of the dearth of detection methods topump themselves up with EPO. The hor-mone, produced mainly in the kidneys,stimulates the body’s production of redblood cells so that the blood carries moreoxygen. People living at high altitudesproduce more EPO naturally to compen-sate for the lower oxygen concentrationin the air. Athletes often take advantageof that trick, training at high altitudes forcompetitions held nearer sea level. Butwhen recombinant EPO, used to treatanemia, became available in the late1980s, it spawned a doping epidemic.

The practice is dangerous. If bloodhas too many red blood cells, it can be-come too viscous for the heart to pumpeffectively. EPO is thought to have playeda role in the deaths of more than a dozen

Dutch and Belgian cyclists who died of sud-den heart attacks in the 1980s, just after EPObecame available in Europe. Despite therisks, EPO’s use was apparently widespreadin the 1990s as scientists raced to figure outhow to detect its use.

The first EPO tests, introduced a decadeago, set a limit for hematocrit, the percentageof red blood cells in the blood. But that test isflawed, as it cannot tell whether an athlete has

used EPO to boost his or her hematocrit to alevel just below the allowed limit.

In 2000, in time for the Olympic Gamesin Sydney, Australia, the IOC introduced acombined blood and urine test for EPO.The blood test measures, among otherthings, the concentration of hemoglobinand the level of reticulocytes—immaturered blood cells—in the blood. Testers lookfor unusually high levels or sudden

changes from previous tests to tipthem off to possible dopers. The testhas one major advantage: It can detectsigns of EPO use weeks after an ath-lete takes it. But because it does notmeasure illegal EPO directly, it can-not prove a doping allegation.

A second method allows testers tospot traces of recombinant EPO di-rectly in urine. Because the recombi-nant version is produced in animalcells, it carries slightly different sug-ars in its side chains than the naturalversion. These differences show up inelectrophoresis, which measures thedistance proteins chug through a gelunder the influence of electricity. Theconcentration of EPO in urine is fair-ly low, however, so the test could befoiled if an athlete takes diuretics orother urine-increasing drugs.

The bottom line is that the currenttests simply don’t cut it. “Athletes aregetting around the EPO tests all thetime,” Catlin says. Officials of theWorld Anti-Doping Agency (WADA)agree. “We need cheaper and moresensitive tests for EPO,” says OlivierRabin, WADA’s scientific director.

WADA is also funding projects totackle an old-fashioned doping tech-nique that the organization claims isback in vogue since the introductionof EPO tests. Called blood doping, itinvolves an athlete either receivingblood transfusions—enriched in redblood cells—from donors, or remov-ing an athlete’s own blood, spinning itto concentrate the red blood cells,then reinfusing it right before compe-tition. Although the techniques don’tinvolve foreign chemicals, they arebanned by sports organizations onsafety grounds.

A growing threatOne of the compounds that BALCOclients are accused of abusing issomething that doesn’t show up inany standard doping tests: humangrowth hormone (hGH). The proteinis part of a biochemical cascade thatspurs muscle buildup and the shed-ding of fat. It’s used legitimately to C

RED

ITS:(

TO

P)

MIK

E H

EW

ITT

/GET

TY

IM

AG

ES;(

BO

TTO

M)

GET

TY

IM

AG

ES


SP

EC

IAL

SE

CT

ION

Take a deep breath. Blood samples will be collected from

athletes in Athens to check for erythropoietin and a range

of other substances.

Banned. Dwain Chambers, who tested positive for the

steroid THG, is barred from the British Olympic team.

treat children who lack the protein and areunusually short. But like EPO and legiti-mate steroids, it too has been hijacked foruse in athletes. Although its effects inhealthy athletes are unclear, doping expertssuspect that its use is widespread—espe-cially because authorities have not yet in-troduced an official test for the compound.

That’s a high priority, however, and sci-entists say they have several tests ready forthe Athens Games. WADA officials are cir-cumspect about whether they will use any ofthe tests for hGH in August. “Athletes knowit is on the banned substances list,” says Rabin, and should expect to be tested.

Detecting hGH is even harder than de-tecting EPO, because it doesn’t have telltalesugars to betray artificial versions. But in alucky break for doping sleuths, the pituitarygland’s production of growth hormone israther messy. The gland makes a mixture ofvariations of the protein as well as proteinfragments. The manufactured version, onthe other hand, is much cleaner, consistingchiefly of one of the heavier versions, sowhen someone shoots up with the recombi-nant protein, the ratio of the different formsis skewed. Endocrinologist Christian Stras-burger of the Charité University Clinics inBerlin and his colleagues at the Medizin-ische Klinik Innenstadt at the University ofMunich have developed an immunoassaythat measures the ratio of the two forms.The test seems extremely reliable, Stras-burger says.

Another group led by Sonksen of St.Thomas’ Hospital has developed a methodto measure the effects of growth hormoneon the production of other proteins, includ-ing insulin-like growth factor-1 (IGF-1) andcollagen. The test is not as clear-cut as thatdeveloped by Strasburger and his col-leagues, but it can detect the effects of hGHweeks after someone has injected it. TheStrasburger method works best 24 to 36hours after injection.

Those who go to the trouble and expense—a month’s dose costs more than$2500—may not be getting their money’sworth. “It looks as though growth hormoneis fool’s gold,” says Ken Ho, an endo-crinologist at the Garvan Institute of Med-ical Research near Sydney. “In the normalperson with normal levels of growth hor-mone, adding extra has not been shown toconfer a benefit.” Yet, Ho says, “at the endof the day, if a 0.01% advantage is the dif-ference between winning and losing,” aminuscule boost from growth hormone—even if it’s purely psychological—mighthelp an athlete to victory.

Tackling this murky question, Ho’sgroup is giving growth hormone to healthyvolunteers both to screen for biochemical

changes that might be picked up in a dop-ing test and to look for performance-enhancing effects. Whether the benefit isreal or not, the hormone is on the list ofbanned substances, and athletes caught us-ing it will forfeit any medals they receivenext month in Athens.

Self-assembled superathletesIn a case that made headlines this summer,doctors described a young boy in Berlinwho seems destined for athletic greatness.The boy was born with a mutation thatturns off the gene for myostatin, which inanimals seems to block the activation ofmuscle stem cells. Mice and cattle thatcarry myostatin mutations have twice asmuch muscle as normal animals. At 4-and-a-half years old, the boy had the physiqueof a mini-bodybuilder and could hold outtwo 3-kg dumbbells with his arms extend-ed, his doctors reportedin the 24 June NewEngland Journal ofMedicine. Some expertsare thrilled: They sug-gest that the mutationcould be exploited as atreatment for muscle-wasting diseases.

But antidoping offi-cials are cringing. Theyfear that gene therapycould soon be the nextfad among athletes.Their nightmare sce-nario is athletes inject-ing a retrovirus bearinga myostatin-blockinggene or another mus-cle-building gene suchas IGF-1 (see sidebaron p. 633). Once thegene is incorporatedinto cells, it begins topump out its products and build muscle,but the illicit source would be extremelydifficult to trace.

Clinical trials of gene therapy for fataldiseases have been fraught with problems,including the death of one volunteer andthe development of leukemia in other pa-tients. But that might not stop some ath-letes. “These [gene therapy] methods re-main extremely risky,” says Rabin. “Buton the other hand we know that some ath-letes are willing to take incredible risks.THG went straight from the test tube tothe athletes, with no proper testing.”

WADA is funding efforts to detect genedoping, either through traces of retrovirusvectors or by spotting indirect effects ofgene boosting, Rabin says. Almost any genedoping would influence a wide range of

other genes, and changes in those might betraceable with ever more sensitive tests thatcan flag gene expression, he says.

As doping sprouts more Medusa-likeheads, authorities may be forced to developpersonalized tests. Ideally, Rabin says, eachathlete would submit a biological “passport”containing highlights of their blood chem-istry. “If we then saw an abnormal change,we could follow it up,” he says. Microarraysthat measure the expression levels of thou-sands of genes at once could betray blipsthat might result from gene doping. Ideally,he says, “in a single drop of blood, we’ll beable to detect changes based on any genesthat are modified.” For now, though, suchtests would cost thousands of dollars perathlete, prohibitively expensive for mosttesting organizations.

Müller doesn’t foresee a breakthrough inantidoping efforts anytime soon. His experi-

ence as a medical scientist in East Germany,where athletes were lavishly funded as aninternational propaganda tool, left him ques-tioning the value of top-level sports. “Weare not dealing here with problems of hu-man existence or survival,” he says. “Theworld will not come to an end if dopers gouncaught.” His motivation, he says, is theexample that elite athletes set for millions ofamateurs: “It’s important for people to beable to understand that you can do amazingthings without doping.”

Unfortunately, many elite athletes don’tbuy that message. “We’re never going toeliminate [doping] completely,” says theUniversity of Ottawa’s Pipe. The contestwill continue, with both sides intent onraising their game to the next level.

–GRETCHEN VOGEL


CR

ED

ITS:T

HE

NEW

EN

GLA

ND

JO

UR

NA

L O

F M

EDIC

INE,

20

04


EC

IAL

SE

CT

ION

Natural boost. A Berlin child who carries a mutation in the myo-

statin gene has had bodybuilder muscles since birth.

When the International Olympic Commit-tee declared at the last minute that “shark-skin” swimsuits were legal for the 2000games in Sydney, Australia, swimmershad just 1 day to decide whether to ditch their tried-and-true swimwear for theflashy new ridged bodysuits that could potentially shave seconds off their times.In a traditionally low-tech sport, choice ofequipment suddenly took on make-or-break significance.

For some of the world’sbest swimmers 4 yearslater, the choice is nolonger theirs. Stars such asMichael Phelps, Inge deBruijn, Lindsay Benko, andGrant Hackett are obligated towear Fastskin, a swimsuit made bySpeedo, their sponsor. The compa-nies TYR and Arena also have swim-mers wearing their versions of thesuit, inspired by sharkskin and intend-ed to cut down on drag, the friction thatslows any body moving through water.On its Web site, Speedo states that Fast-skin reduces drag up to 4%. Other com-panies make similar claims.

But whether the suits work is hotly de-bated. Critics dispute the explanations ofhow the suits reduce drag. “The claimsthat the manufacturers make about the sci-ence behind what these suits do are outra-geously ignorant,” asserts Brent Rushall, aprofessor of exercise and nutrition at SanDiego State University in California.

The concept itself is not unreasonable.Sharkskin is textured with ridges, knownas “riblets,” that reduce the amount of skinsurface area that comes in contact withwater. That allows sharks to glide throughwater with much less friction than onewould expect. In 1987, the hull of anAmerica’s Cup sailboat was textured tomimic the skin of pelagic sharks; it wonthe race and was so fast that the texturingwas outlawed. The advantage made sense:Pelagic sharks cruise long distances atspeeds approaching those of sailboats,around 15 knots, or 28 kilometers perhour, so the benefit from the riblets wastransferred from beast to boat.

Manufacturers say that the sharkskinsuits are grounded in sound science. Thesuits are designed to lower two types ofdrag: skin friction and pressure drag. The

trick for reducing skin friction is makingthe riblets just the right size. If they aretoo big or too small, water floods the val-leys, creating more surface area than astandard suit has, says Barry Bixler, anaerospace engineer who models fluid dy-namics for Speedo. Pressure drag, mean-while, occurs when water flowing over ir-regular shapes, such as eyebrows, chins,and breasts, peels off in sheets, creatingsuction. Imagine water flowing over abeach ball until it reaches a critical angle

and falls away, creating a vacuum thatsucks at the ball. To reduce this ef-

fect, the Speedo suit is dimpled,like a golf ball, in strategic

places. The indenta-tions set up minia-

ture eddies; tur-bulent water

is more

likely tostick tothe swim-mer than issmooth-flow-ing water. The suitsare designed to keep“sticky” water continu-ously flowing over the body,thus reducing pressure drag.

According to Bixler, Speedo’smodels are based on body scans oftwo Olympic medalists, a man and awoman (Speedo will not reveal theirnames), in a streamlined diving position.Mannequins of the swimmers match themodel’s calculated drag predictions towithin a few percent, Bixler says. Howev-er, neither the mannequins nor the com-puter-generated swimmers, the basis forcalculations of passive drag, stroke orkick the way a human would. “It’s diffi-cult to model that,” says Bixler. However,he says, “we know that if passive drag isless, active drag will also be less.” BySpeedo’s calculations, the bodysuits re-

duce passive drag 4% in men and 3% inwomen, whose hips and breasts tend to in-crease pressure drag.

One skeptic is Huub Toussaint, a bio-mechanist at the Free University in Am-sterdam. The manufacturers “take theswimmer and put him underwater in aflow tank with [a] steady flow, … andthey ignore that in reality, the swimmer ismoving.” In Sport Biomechanics in 2003,Toussaint and colleagues comparedswimmers propelling through a poolwearing training suits with swimmerswearing sharkskin suits. The study foundno significant reduction in drag from thesharkskin suits.

Rushall, meanwhile, argues that theflow of water over a boat’s hull is nothinglike that over a swimmer. The fastest hu-mans thrash through water at a maximumspeed of about 7 kilometers per hour, aquarter of the speed of a pelagic shark or asailboat. “Whatever animal [the swimsuitmanufacturers] are trying to reflect is notgoing to be swimming as slowly as a hu-man,” says Rushall. A texture that lessensfriction at shark speeds may even increasefriction at human speeds, he says. Rushallclaims that a Japanese swimsuit maker,which he declined to name, had tested a

sharkskin-style surface material andfound that it had reduced friction only at very specific angles of entryto the water. The company conclud-ed that any reduction of drag on the

irregular bumps and indenta-tions of the human body

would be minimal anddropped the project.

Independent testsare hard to carry

out. Bixler con-f irms that the

suits reduceboth frictionand pressure

drag but saysthat Speedo has

barred him from publishing his modelingdata on the suits. Speedo and other manu-facturers contacted by Science say that thedata are trade secrets.

The uncertainty has led some to focuson the test that really matters: Olympictimes. As a result of improvements intraining, nutrition, and technique, swim-mers steadily improve in a statisticallypredictable way, says Joel Stager, a physi-ologist at Indiana University, Blooming-

N E W S

Do Pool Sharks Swim Faster?New swimsuits with tiny ridges modeled on sharkskin are all the rage.Experts are split, though, about whether the high-tech suits reduce drag


SP

EC

IAL

SE

CT

ION

CR

ED

IT:B

AR

RY

BIX

LER

/SPEED

O


ton. The data wobble when performanceenhancers such as doping or technologi-cal breakthroughs enter the picture. Afteranalyzing swim times in the SydneyOlympics, Stager says, “If the suitsworked, we shouldn’t be able to predictperformance. But lo and behold, we werepretty much dead-on except for twoevents.” Of those two, the men’s 100-meterbreast stroke was faster than predicted,with all the athletes wearing conventionalswimsuits. The other event, the women’s200-meter backstroke, was slower thanpredicted—with everyone wearing thesharkskin suits.

Next month’s Olympics in Athens willprovide a high-prof ile venue for thesharkskin suits to sink or swim. Manufac-turers have had 4 years to hone theirproduct, and more athletes will havepracticed with them and have their ownspecially fitted versions.

Effective or not, the suits are catch-ing on. “I haven’t seen a freestyler racein anything but a Fastskin for as long asI can remember,” says Kevin Swander,the 2004 Big Ten champion, who com-peted in the Olympic trials in LongBeach, California, in July. Swanderlikes the bodysuits and has worn one at

every single meet of his college career.“I don’t know if it reduces drag. I likethe way it feels,” he says. “The biggestthing with these suits is the mental aspect. The suit supposedly makes youswim faster, so [when] you’re wearingthem—you swim faster.”

Stager, for his part, is wistful for thedays when anyone with a $30 swimsuitcould compete. When modern Olympiansput their full-body sharkskin suits on, “it’slike they become invisible,” he says. “Theylose their identity.” One thing has notchanged, at least: In the pool, speed alwaystrumps appearance. –KIM KRIEGER


CR

ED

IT:B

ET

TM

AN

N/C

OR

BIS


EC

IAL

SE

CT

ION

In 1968, a Kenyan runner named KipKeino emerged as a shining star of theMexico City summer Olympics, setting aworld record in the 1500-meter race. Yearafter year Keino’s success has been fol-lowed by equally dazzling feats by hiscompatriots: Kenyan men now hold worldrecords in the 3000-meter track race, the15-, 20-, and 25-kilometer road races, thehalf-marathon, and the marathon. Kenyanmen have won 13 of the last 14 Bostonmarathons. Kenyan women are alsorising fast: They hold half of the top10 marathon times and world recordsin 20-, 25-, and 30-km track races.What is even more remarkable is thatmost of these athletes come from asmall area in Kenya’s Rift Valley,from a group of tribes called theKalenjin who number little more than3 million people.

Theories abound about whatKenya-born writer and runner JohnManners calls “the greatest geograph-ical concentration of achievement inthe annals of sport.” Is it the high alti-tude that fosters big lungs and effi-cient oxygen use? Is it their maize-based diet? Or the fact that manychildren run to school? A gruelingtraining regimen, perhaps?

Such questions have inspired ahandful of researchers to try to definethe Kenyan magic. Meanwhile, scien-tists are unraveling why athletes whoseancestors come from the other side ofthe continent—West Africa—haveemerged as the world’s fastest sprinters.

Fuel economyLeading the charge in penetrating theKenyan mystique has been Bengt Saltin, aSwedish physiologist who heads theCopenhagen Muscle Research Centre inDenmark. In the 1990s, Saltin’s group be-gan comparing Kenyan and Scandinavianrunners by scrutinizing their physiologicalmakeups and assessing the “trainability”of novice runners in both countries.

A decade later, the scientists have ruled

out most of the popular explanations forKenyans’ domination of running. Altitudeis not the key to the riddle, they have found,because there’s no difference betweenKenyans and Scandinavians in their capacity to consume oxygen. And theKenyan diet is on the low side for essentialamino acids and some vitamins as well asfat, says Dirk Christensen of the Copen-hagen center: “In spite of the diet, they per-form at high level.” The running-to-schoolhypothesis was demolished as well: Kenyanchildren aren’t any more physically activethan their Danish peers. Do Kenyans tryharder? The researchers found that the

Danes actually pushed themselves hard-er on a treadmill test, reaching highermaximum heart rates.

An important clue is the ability ofKenyans to resist fatigue longer. Lac-tate, generated by tired, oxygen-deprived muscles, accumulates moreslowly in their blood. Comparisons oflactate levels have suggested to Saltin’sgroup that Kenyan runners squeezeabout 10% more mileage from the sameoxygen intake than Europeans can.

Just as more aerodynamic cars getbetter gas mileage, the Kenyan buildhelps explain their fuel efficiency. Arecent British TV documentary de-scribed the Kalenjin as possessing“birdlike legs, very long levers that arevery, very thin [on which they] bounceand skip” along.

Saltin’s group has quantified thisobservation. Compared with Danes,the thinner calves of Kenyans have, onaverage, 400 grams less flesh in eachlower leg. The farther a weight is fromthe center of gravity, the more energyit takes to move it. Fifty grams added

N E W S

Peering Under the Hood of Africa’s RunnersKenyans dominate endurance running, and West Africans excel as sprinters. With a physiological explanation in hand, researchers are nowprobing the genetics of this geographic mastery

Running revolutionary. Kenya’s Kip Keino in 1972.


to the ankle will increase oxygen con-sumption by 1%, Saltin’s team calculates.For the Kenyans, that translates into an 8%energy savings to run a kilometer. “Wehave solved the main problem,” declaresHenrik Larsen of the Copenhagen center.“Kenyans are more efficient because ittakes less energy to swing their limbs.”Other scientists say the jury is still out onthe Kenyan question. But “I think Saltin isprobably the mostcorrect that anyoneis at the moment,”says physiologistKathryn Myburghof the University of Stellenbosch inSouth Africa, who isexploring the role ofKenyans’ training.

However, slimlower legs are notthe whole story.Kenyan runners alsohave a higher con-centration of an en-zyme in skeletalmuscle that spurshigh lactate turn-over and low lactateproduction. Saltinsays that this resultsin an “extraordinar-ily high” capacityfor fatty acid oxidation, which helpswring more energy out of the muscles’biochemical reactions. Because intensetraining alters the body’s biochemistry,Saltin says that he can’t say for surewhether the ezyme levelsare due to genes or train-ing. But he adds, “I thinkit’s genetic.”

Research in SouthAfrica jibes with theCopenhagen group’s find-ings. A team led by exer-cise physiologist AdeleWeston of the Universityof Sydney, Australia,compared black SouthAfricans, whose runningstrengths are similar tothose of Kenyans, withwhite runners. The twogroups had similar VO2max values—that is,when putting out maxi-mum effort, they used upthe same amount of oxy-gen per kilogram of bodyweight per minute. Butthe black runners weremore eff icient in their

oxygen consumption, lasting on a tread-mill at maximum speed for twice as longas the whites. As with the Kenyans, theblack South African runners accumulatedless lactate and had higher levels of keymuscle enzymes.

A little more twitchyWhereas East Africans dominate long-distance running, West Africans have surged

to the fore in short-distance events. Little re-search has been done on West Africans, butthere’s powerful circumstantial evidence forsome physical advantages, as presented byJon Entine in his book Taboo: Why Black

Athletes Dominate Sportsand Why We’re Afraid toTalk About It. Athletes ofprimarily West African de-scent—which includes themajority of U.S. blacks—hold all but six of the 500best times in the 100-meterrace, “the purest measure

of running speed,” says Entine, whose bookset off a broad debate on the subject.

Various studies have shown that WestAfrican athletes have denser bones, lessbody fat, narrower hips, thicker thighs,longer legs, and lighter calves than whites.But the differences between East and WestAfricans are even more striking. The fa-bled Kenyan runners are small, thin, andtend to weigh between 50 and 60 kilo-

grams, whereasWest African ath-letes are taller and agood 30 kilogramsheavier, says Timo-thy Noakes, a prom-inent exercise phy-siologist and re-searcher at the Uni-versity of CapeTown.

The differencesdon’t stop withbody shape; there isalso evidence of adifference in thetypes of musclefibers that predomi-nate. Scientists havedivided skeletalmuscles into twobasic groups de-pending on theircontractile speed:

type I, or slow-twitch muscles, and type II,fast-twitch muscles. There are two kindsof the latter: type IIa, intermediate be-tween fast and slow; and type IIb, whichare superfast-twitch. Endurance runnerstend to have mostly type I fibers, whichhave denser capillary networks and arepacked with more mitochondria. Sprinters,on the other hand, have mostly type IIfibers, which hold lots of sugar as well asenzymes that burn fuel in the absence ofoxygen. In the 1980s, Claude Bouchard’steam at Quebec’s Laval University tookneedle biopsies from the thigh muscles ofwhite French Canadian and black WestAfrican students. They found that theAfricans averaged significantly more fast-twitch muscle fibers—67.5%—than theFrench Canadians, who averaged 59%.Endurance runners have up to 90% ormore slow-twitch fibers, Saltin reports.

Bouchard, now at Louisiana State Uni-versity in Baton Rouge, says his teamlooked at two enzymes that are markersfor oxidative metabolism and found higheractivity of both in the West Africans,meaning they could generate more ATP,the energy currency of the cell, in the ab-sence of oxygen. The study suggests thatin West Africa there may be a larger pool


SP

EC

IAL

SE

CT

ION

Triumph of fast twitch. Carl Lewis, a U.S. sprinter with West African roots, winning the 400-meter

race at the 1992 Olympics in Barcelona.

5.0

4.2

13.3

3.313.3

12.5

48.3

Asia

Kenya

South AmericaAfrica excl. Kenya (includes U.S. blacks)

North America

Europe

Australia

1986

2.5

29.2

11.7

0.8

55.8

2003

Out of East Africa. In less

than 2 decades, Kenyans

came to dominate the top 20

performances in six races

ranging from 800 meters to

the marathon.

CR

ED

ITS:(

TO

P)

JEA

N-Y

VES R

USZ

NIE

WSK

I/T

EM

PSPO

;(B

OT

TO

M)

HEN

RIK

B.L

AR

SEN

,J. C

OM

P. B

IOC

HEM

. PH

YSIO

L.

of people “with elevated levels of what ittakes to perform anaerobically at veryhigh power output,” says Bouchard.

Although training can transform super-fast-twitch type IIb fibers into the hybridtype IIa, it is unlikely to cause slow- andfast-twitch fibers to exchange identities.Myburgh says there is evidence that, withextremely intensive long-distance training,fast IIa fibers can change to slow type Ifibers. So far, however, there is no evi-dence that slow-twitch f ibers can beturned into fast-twitch ones. As an athleteputs on muscle mass through training, newfibers are not created, but existing fibersbecome bigger.

Running ACEsThe differences in physique and musclemakeup that underlie the dominance ofKenyan endurance runners and WestAfrican sprinters doubtless have a stronggenetic component. But researchers areonly just getting off the starting mark inthe search for genes that influence run-ning performance. Bouchard’s group, forexample, is collecting DNA samplesfrom 400 runners and other top en-durance athletes from the United Statesand Europe, but he says they haven’tspotted any running genes yet.

There are a couple of intriguing possi-bilities, though. In 1999, a team headed byKathryn North of the Children’s Hospitalat Westmead in Australia described twoversions of a gene that affects productionof α-actinin-3, a protein found only infast-twitch muscles. They found the lessefficient version of the gene—which re-sults in poorer energy conversion—in 18%of the members of a group of Caucasians.In 2003, North’s group reported in theAmerican Journal of Human Genetics thatonly 6% of a group of sprinters had thegene defect; 26% of endurance runnershad it. The authors surmise that α-actinin-3 helps muscles generate “forceful con-tractions at high velocity.”

Alejandro Lucia Mulas of the Euro-pean University in Madrid is taking DNAsamples from Eritrean runners to exploreanother candidate: different versions of thegene for angiotensin-converting enzyme(ACE). Lucia says the less active version,or I allele, of this gene is associated withless muscle, less fluid retention, and morerelaxed blood vessels—which would en-hance oxygen uptake—and appears to bemore prevalent in endurance runners.

And in Scotland, sports physiologist Yannis Pitsiladis has launched a major on-slaught on the Kenyans’ secrets with the

International Centre for East African Run-ning Science. Headquartered at the Uni-versity of Glasgow, the virtual center willbring together research on demography,diet, and socioeconomic factors as well as

genes. Pitsiladis says he has spent the last3 years in East Africa collecting DNAsamples from their “living legends” andnow has DNA from 404 Kenyan and 113Ethiopian athletes. His team has found ahigher prevalence of the I allele for theACE enzyme in male marathoners com-pared with men from the general Ethiopianpopulation. But Pitsiladis thinks his num-bers may lack significance given the vari-ability of the trait in African populations.“At the moment there is no evidence” thatEast Africans have a genetic advantage inrunning, he says.

None of the data negate the importanceof cultural habits and training. But as En-tine quotes anthropologist and sports sci-ence expert Robert Malina, who is retiredfrom Michigan State University, “Differ-ences among athletes of elite caliber are sosmall that if you have an advantage thatmight be genetically based … it might bevery, very significant.”

Next month’s Olympic games inAthens should demonstrate yet again thatWest African runners are built for speedand Kenyans built to endure.

–CONSTANCE HOLDEN



EC

IAL

SE

CT

ION

CR

ED

IT:T

HO

MA

S N

OLA

N

When the U.K.’s Paula Radcliffe ran theLondon Marathon last year in just under135 minutes, shaving almost 2 minutesoff the record she set in 2002, the whis-pers started again: Are women improvingtheir performance so quickly that one daythey may compete on the same trackswith men?

Expert opinion suggests that day will re-main elusive—as long as women retain fe-male bodies. The gap between the sexes ap-pears to have plateaued, with women per-forming at about 90% of male levels. Apartfrom the marathon, “world records forwomen have been absolutely static” formore than a decade, notes Kenya-born jour-nalist and running expert John Manners.

That plateau wasn’t evident 12 yearsago. In a letter to Nature published on 2January 1992, titled, provocatively, “Willwomen soon outrun men?,” Brian L. Whippand Susan Ward of the University of Cali-fornia, Los Angeles, looked at the worldrecords of five standard Olympic runningevents, from the 200-meter dash to the 26-mile (42-kilometer) marathon, from the1920s through 1990. They found that

women were improving their times at dou-ble the rate of men in the short distancesand were narrowing the gap even faster inthe marathon, in which record-keeping forwomen only started in 1955. “The gap isprogressively closing,” the authors wrote.At that rate, they projected that marathontimes could converge by 1998 and thatgender differences in all races could disap-pear by 2050. In 1996, a poll by U.S. Newsand World Report reported that two-thirdsof Americans believed that “the day iscoming when top female athletes will beattop males.”

Absent unforeseen genetic or hormonalinterventions, men, it seems, will maintainan advantage. That’s due largely to theirsteady supply of a performance drug thatwill never be banned: endogenous testos-terone, which boosts muscle power andoxygen capacity. The typical young manhas a maximum oxygen use capacity, orVO2 max, of about 3.5 liters per minute,compared with 2 liters for a woman, saysphysiologist Stephen Seiler of the Instituteof Health and Sport at Agder College inKristiansand, Norway. Although individual

N E W S

An Everlasting Gender Gap?For a while, female runners were closing in on their male counterparts.Now they’re barely keeping the guys’ taillights in sight

Road test. Masks monitored Kenyans’ oxygen use.


CR

ED

ITS:(

TO

P)

AD

APT

ED

FR

OM

S.S

EIL

ER

;(B

OT

TO

M)

ST

U F

OR

ST

ER

/GET

TY

IM

AG

ES

levels of testosterone vary widely, malestend to have at least 10 times as much ofthe stuff as women. The hormone stimu-lates the creation of red blood cells, whichmeans that men’s blood holds about 10%more of the oxygen-carrying proteinhemoglobin. But oxygen is not the wholestory. Kirk Cureton of the University ofGeorgia School of Health and Human Per-formance in Athens compared the per-formance of male and female athletes onan exercise bicycle after scientists hadwithdrawn blood, leaving the subjects withequal amounts of hemoglobin in circula-tion. That reduced but did not eliminate thesex difference in VO2 max, indicating thatother factors, particularly musculature,play into the difference.

Men have more muscle and largerhearts in relation to body size, says DirkChristensen, an exercise physiologist at theUniversity of Copenhagen. This affects aer-obic capacity: He says that a trainedwoman’s heart can pump out the same vol-ume of blood as a man’s can, but it has towork much harder to do so.

Because testosterone spurs growth ofmuscle tissue, it also affects anaerobic ca-pacity—the ability to produce energyquickly without oxygen—which givesmales an edge in sprinting as well. Theprimary energy for the intense bursts ofpower required for sprints is generatedanaerobically, explains retired MichiganState University anthropologist and sportsexper t Rober t Malina. Indeed, afterlaunching themselves for a 10-secondsprint, some athletes “don’t take anotherbreath till it’s all over,” he says. (En-durance running in contrast relies almostexclusively on aerobic energy.) Moremuscle means more of the two mainanaerobic energy sources: phosphocrea-

tine and glucose.Recent records sup-

po r t t he gende r-gapplateau, Seiler says. A fewyears ago, he and writerSteve Sailor analyzed results from Olympicgames and world champi-onships of the Interna-tional Association of Ath-letics Federations between1952 and 1996, selectingevents in which men andwomen ran under thesame conditions. Theyfound that if the marathon— w h i c h w a s n ’ t a nOlympic event for womenuntil 1984—were exclud-ed, the mean performancegap for running events in-

creased from 11% in the mid-’80s to 12% inthe mid-’90s. They also observed that men’sworld records were broken far more often inthe ’90s than women’s—largely due to theextraordinary performance of East Africanrunners (see p. 637).

Seiler recently updated these numbersfor Science. He reports that from the worldrecords in the eight main running eventsfrom 100 meters to the marathon, sevensuggest an increasing gender gap. The

marathon is the exception: The gap hasnarrowed from 11.9% to 8.4%, thanks toRadcliffe’s new record. Seiler says the cur-rent average gap is now 11.01%, up from10.4% in 1989. In short, he says, “at thehighest levels of performance, the gendergap in running performance has actuallywidened over the last 20 years.”

Much of the female record is cloudedby drug use, especially the records set inthe 1970s and ‘80s by Eastern Europeanwomen that have never been bested. In1984, 38 women, mostly from the EastBloc, ran 1500 meters in under 4.05 min-utes, according to Jon Entine in his bookTaboo. In 1991, only nine did.

Although the impressive gains in femalemarathon performance have suggested tosome observers that women have greaterendurance than men, physiologist HenrikLarsen of the Copenhagen Muscle Re-search Centre says that’s not so: “Womenhad not developed long distance; that’swhy the improvement is much greater onthe marathon. We don’t see any higher ox-idative capacity in women.” Exercise phys-iologist Timothy Noakes of the Universityof Cape Town, South Africa, agrees. Asmaller body frame gives women an edgeon endurance, he says, but men can run10% faster even when the difference inbody size is controlled for.

Whipp says he’s still keeping anopen mind on the subject of male-female competition. He told Sci-ence that he and colleagues are cur-rently working to extend theiranalysis of world running recordsto 2003. His team has looked for “apattern of response that would sug-gest a physiological limit,” he says,but so far has found none. “There isno evidence at the beginning of the21st century that the human athletehas reached the limit of [his or her]potential,” Whipp says.

But that appears to be a minorityview. “We are approaching the lim-its of human performance in a lotof the one-dimensional events likethe 100-meter sprint or marathon,”says Seiler. “Records will continueto be broken, but the price is ex-tremely high. And the percentageof the population that has the ge-netic potential to excel at this levelis infinitesimal.” As for the gendergap in running, he defers to Nor-way’s marathon queen Grete Waitz,setter of world records in the 1970sand ’80s, who said: “As long aswomen are women, I don’t thinkthey will surpass men.”

–CONSTANCE HOLDEN


80

70

60

50

40

30

20

10

01970s 1980s mid-1990s

Males

Females

%

Percentage of all track-and-field medals won by Eastern bloc athletes.

Fall of eastern superwomen. Drug use helped inflate female per-

formance before the fall of communism.

Runaway record holder. Paula Radcliffe keeps carving

minutes off the marathon.

SP

EC

IAL

SE

CT

ION

When Natalia Yurchenko introduced a newvault at the 1983 World Championships inBudapest, it helped her win a gold medal.The move is fiendishly difficult: Yurchenkowould sprint 20 meters, cartwheel, landbackward on a springboard, and launch her-self into the air. Arching her back, shewould reach for a padded apparatus called ahorse and then propel herself into the airagain, somersault one-and-a-half times,twist, and land facing backward.

Following that dizzying lead, gym-nasts set out to conquer the Yurchenkovault—and injuries mounted. “Every sin-gle country had problems,” recallsWilliam Sands, head of sport biomechan-ics and engineering at the U.S. OlympicTraining Center in Colorado Springs,Colorado. The vault became notorious in1988, when 15-year-old Julissa Gómezbroke her neck while attempting it at theWorld Sports Fair in Japan. She was par-alyzed, fell into a coma, and later died ofcomplications. Within a year, the U.S.Gymnastics Federation, predecessor ofUSA Gymnastics, banned the move atlevels below Olympic competition.

Deaths are rare in gymnastics. That’snot the case for injuries, which Sandscalls “the most pressing and serious prob-lem faced by contemporary gymnastics.”Compared with other kinds of athletics,gymnastics stands out for its singularcombination of bone-jarring impacts, in-tense training, young age, and ever-more-demanding skills. Similar injuries afflictboth sexes, but more girls participate andthey start training younger. Ex-perts fear that elite teens andpreteens, by pushing their bod-ies to the limits, might be rais-ing their risk of osteoarthritisand other health problems laterin life. “Kids may be sustain-ing injuries that will be withthem for a long time,” saysLyle Micheli, an orthopedicsurgeon at Children’s Hospitalin Boston.

There’s no question thatgymnastics is a punishingsport. Long hours of practiceexact considerable wear andtear. An elite gymnast trains 25to 40 hours a week, typicallyexecuting more than 250,000

“skills” a year. In the 1970s and ’80s, elitegymnasts began competing at youngerages. Although that trend has stabilized,preteens are subjecting their bodies tosprains, fractures, and sometime even de-formities of growing bones.

Gymnastics is also unusual in that it hasbecome more demanding as judges revisethe “Code of Points” used to score nationaland international competitions. In 1996, es-tablished routines became worth fewer

points, so gymnasts who wanted to outscorethe competition had to swing higher into theair, execute more twists and somersaults,and otherwise raise their game. For a fewyears after that, the U.S. national team hadproblems with a serious kind of knee injury,a tear of the anterior cruciate ligament. Alsoraising the risk of injury, equipment hasbeen modified for higher performance. Forexample, gymnasts can jump higher fromnew vaulting tables with more spring, al-lowing more air time for acrobatics.

Sports scientists say it is hard to naildown the health risks of competitive gym-

nastics for children. For starters, there isno reporting system for gymnastics in-juries. In the one existing system, run bythe National Collegiate Athletic Associ-ation, gymnastics usually ranks in thetop three sports for injuries, behind foot-ball and hockey. But that doesn’t shedlight on the private clubs that train mostgymnasts. Coaches of elite gymnasts of-ten may be too busy to let scientists infor a close look, says Patrick O’Connor,an exercise scientist at the University ofGeorgia in Athens.

With fewer than 200 elite gymnasts inthe United States, most research is doneon less-skilled athletes. These studiesback the impression that injuries are com-mon. In a 3-year study of 79 female gym-nasts aged 7 to 18, Dennis Caine, asports-injury epidemiologist at WesternWashington University in Bellingham,found that 60 girls suffered a total of 192sprains, strains, and other injuries. Thestudy also showed that the risk of injurywas significantly higher for advancedgymnasts. That makes sense, because as

gymnasts get serious, they put inlonger hours, work harder, andattempt more difficult routines.

Caine and others have shownthat the most commonly injuredbody parts in boys and girls in-clude the lower back, shoulder,and ankle. Wrist pain is especiallyprevalent, says John DiFiori, chiefof sports medicine and a teamphysician at the University ofCalifornia, Los Angeles. In somecases, the distal end of the fore-arm bones can be damaged, stunt-ing the radius relative to the ulna.“It can be career-ending for somekids, because weight-bearing istoo painful,” says Caine. Suchnagging injuries can continue to


CR

ED

ITS:(

TO

P)

MIK

E P

OW

ELL

/GET

TY

IM

AG

ES;

(BO

TTO

M)

AD

APT

ED

FR

OM

BIL

L SA

ND

S/T

ECH

NIQ

UE

MA

GA

ZIN

ET E S T I N G H U M A N L I M I T S S

PE

CIA

LS

EC

TIO

N

N E W S

Graceful, Beautiful, and PerilousAs gymnastics routines grow ever trickier, experts worry that children arebeing pushed beyond their limits—and are paying with their health

The agony of victory. Kerri Strug helped the U.S.

team win gold by vaulting with an injured ankle.

200

195

190

185

180

175

170

165

Sco

re

1983 1985 1987 1989 1991 1993 1995 1997

Year

oresChampionship Scom o h Sp pNCAA National CC A a n C oWomenW e ’s teamsm

High performance. Top gymnasts are scoring higher and competing hard-

er—raising the risk of injury.


plague gymnasts in college, although he saysit’s unclear whether they flare up later in life.

For girls, another concern is delayedgrowth and sexual maturation. Elite femalegymnasts tend to grow more slowly and gothrough puberty later than other girls.Robert Malina, a retired auxologist in BayCity, Texas, doesn’t think intense trainingcan be blamed; he says a confounding fac-tor is that larger girls who mature earliertend to drop out from competition. How-ever, Caine points to several case and co-hort studies that indicate atleast a temporary halt ingrowth of some top-levelgymnasts, likely due to in-tense training and poor nu-trition, he says. When thegymnasts lightened theirload, or retired, theirgrowth rate accelerated.

Many of these issuescame to a boil during the1992 Barcelona Olympics,when European newspapersran stories about the injuriesof young female athletesand the extreme trainingthey had undergone. Thatspurred a research projectthat many sports scientistscall exceptional in its depthand quality.

The Federal Institute for Sport Science inBonn, Germany, asked Gert-Peter Brügge-mann, then director of the Institute for Athlet-ics and Gymnastics at the German Sport Uni-versity Cologne, to study the effects of high-level performance on gymnasts. His team ex-amined various training regimens, usingrecords from the once top-secret facilities ofthe former East Germany. Compared withWestern gymnasts, the East Germans wereput through more frequent repetitions of skillsets and more difficult maneuvers and gotless rest. In total, their growing bodies hadendure a longer, harder pounding comparedwith those of West German athletes.

Those loads had severe consequences.After reviewing archived x-rays and re-examining 42 women and 26 men who oncehad been elite gymnasts, Brüggemann’sgroup found a much higher injury rateamong the East German team than in 23West Germans who had competed between1968 and 1985. Mild deformities and ab-normalities of the spine were more thantwice as common in the East Germans,Brüggemann and colleague Hartmut Krahl,then of the Alfried Krupp Krankenhaus inEssen, reported in 2000 in Belastungen undRisiken im weiblichen Kunstturnen (Loadand Risks in Female Gymnastics).

Seeking to prevent such injuries, the

group also carried out a 4-year prospectivestudy of 135 young elite gymnasts on Ger-man national squads. Working with high-speed video cameras and force-measuring de-vices in the apparatus and landing mats, theyanalyzed roughly 100 exercises and measuredmechanical loads exerted on the body overtime. About half the spinal deformities couldbe explained by the amount of loading. Theyalso discovered that the severity tended to beworse among those with weaker muscles andconnective tissue. Stronger muscles absorb

the shock of impacts and bad landings, pro-tecting joints and the spine.

Based on those findings, Brüggemann’steam devised a healthier training regimen inwhich girls spent less time learning fancyroutines, instead logging more hours in theweight room. Far fewer gymnasts sufferedpain or injury during the first 3 years of thestrength-training program, they found, withankle injuries alone falling more than 50%.One beneficiary of the program is Brügge-mann’s daughter Lisa, who competed in the1999 and 2001 World Championships andwill be heading to Athens on the Germanteam. “As long as she’s in this [training] sys-tem, I feel comfortable,” Brüggemann says.

A similar approach has helped the U.S.national team, which has suffered fewerknee injuries since putting a greater empha-sis on strength training in 2000. “We workon fitness and health rather than just skills,”says team physician Lawrence Nassar ofMichigan State University in East Lansing.Fitness and health are the best predictors ofwho will make the Olympic squad, he says.

Jill McNitt-Gray, who studies biomechan-ics at the University of Southern California inLos Angeles, is also working to make gym-nastics safer. “I got into this because I was agymnast and a coach,” she says. “I saw toomany athletes getting hurt, and I thought

there must be a better way.” In her lab, McNitt-Gray takes high-speed video of gym-nasts, measures forces and muscle move-ments, and creates computerized stick mod-els to test the effects of modifying the moves.The gymnasts are outfitted with tiny sensorsto track their motions and muscle use. Thissystem can be used to warn trainers if thegymnasts are nearing their limits, McNitt-Gray says: “If they can’t control [the mo-tions], they’re more likely to get injured.”

Also on the agenda is improving theequipment. By filmingthe vault board andspring floor with a high-speed camera, Sandsand his colleaguesfound that it wobblesunderfoot before heav-ing the gymnast up. “Itlooks like the gymnasthas landed on a water-bed,” he says. That’s partof what makes theequipment risky, be-cause gymnasts mustcope with boards thatoften compress uneven-ly and unpredictably.Boards have improvedover the years but arestill not good enough,says Sands. Similarly,

the characteristics of the safety mats thatgymnasts land on depend on their con-struction, age, and other factors. Brügge-mann and his colleagues have recom-mended that safety mats be stiffened forbetter stability during landing.

Most competitive gymnasts cannot avoidan occasional injury, but little is knownabout the long-term damage to their health.Anecdotal evidence suggests that elite gym-nasts risk developing osteoarthritis andchronic musculoskeletal pain. “In mostsports, if you compete at a high level you’regoing to carry some baggage with you for along time,” Sands says. The few studies thathave tested this hypothesis have yieldedconflicting results.

Experts insist that gymnastics can bemade safer without diminishing its eleganceand power. “It is not as much an art as peo-ple like to think. We’ve got plenty of scienceto do and plenty of well-understood tools tofind out what we need to know,” says Sands.“The problem is lack of money, courage,and commitment.” Perhaps the most impor-tant message is “not to cross the pain thresh-old,” says Caine. “The old adage ‘No pain,no gain’ is inappropriate when it comes tokids.” In the pursuit of Olympic gold, how-ever, that message can easily get lost.

–ERIK STOKSTAD CR

ED

IT:D

ON

ALD

MIR

ALL

E/G

ET

TY

IM

AG

ES


SP

EC

IAL

SE

CT

ION

Impact. The extreme forces in gymnastics, especially landings, can take a toll.

DAVIS, CALIFORNIA—Mont Hubbard handlesa weighty discus the way an anthropologistmight examine a 10,000-year-old skull,turning it gingerly in his fingertips. “Itrolls out of the thrower’s hands spinningthis way,” he says, turning the shiny blue“artifact,” which might break your foot ifit fell on you, in ultraslow motion. Withhis gray beard and gangly build, the 61-year-old looks nothing like an elite athleteas he rocks back in his well-worn chair inan office cluttered with baseball bats, Fris-bees, and thick binders labeled “shot put,”“fly casting,” and “pole vault.” Yet Hub-bard knows how to hurl a javelin as far asit can go, guide a speeding bobsled mosteff iciently through a hairpin turn, andknock the longest home run.

Whereas athletes rely on countless hoursof practice to master their sports, Hubbarddeduces winning ways from his prodigiousunderstanding of mechanics. “Almost allsports is mechanics,” Hubbard says. “Al-ways there are things that are moving, andalmost always the motion is central. Youwant to know how to get something to movein a certain way, or how to get it to go far-ther.” When it comes to analyzing andoptimizing motion, Hubbard and hisgroup at the University of California,Davis, got game. Employing basicphysics and sophisticated mathematicaltools, they calculate what athletes try tofeel: the best way to move.

Hubbard has studied, among otherthings, the mechanics of the high jumpand the pole vault, the aerodynamics ofthe discus and the shot put, the flight ofpunted footballs, the bend of curveballs,the flexing of fishing rods and racehorseforelimbs, the swiveling of skateboards,and the rolling of basketballs aroundrims. In the 1980s, he developed a sys-tem to help javelin throwers launch theirspears at just the right angles. In the ’90s,he built a bobsled simulator for U.S.Olympians. Currently, Hubbard and hisgroup are analyzing ski jumping,women’s gymnastics, trampoline, bungeejumping, and Frisbee flight. Last year, hekicked up controversy by arguing that abaseball batter can knock a slower-moving curveball farther than a fastball(see sidebar on p. 644).

“Hubbard is perhaps the leading fig-ure in determining optimal strategies in

sports,” says William Stronge, a mechanicalengineer at the University of Cambridge,U.K., who has studied the dynamics ofbouncing balls. Neville de Mestre, an ap-plied mathematician at Bond University inGold Coast, Australia, who has done re-search on fluid dynamics in sports, saysHubbard is exceptionally eager to put hisideas to the test. “Many people talk aboutsimulations,” he says, “but Mont actuallybuilds experiments to see how it works.”

Hubbard’s work may be of gold-medalquality to his peers, but he has had a hardtime winning over coaches and athletes.“Most people in sports don’t think it’s use-ful,” he admits. That may change, however,as Hubbard and other practitioners of“sports engineering” develop more powerfuland realistic analyses that can better predictthe keys to victory.

For the fun of itWalking into Hubbard’s Sports Biomechan-ics Laboratory on the Davis campus is likestepping into a strange high-tech equipmentlocker, minus the stench of sweaty socks.On the back wall of the windowless room

hangs a rack of javelins, their menacingpoints ready to perforate the air. Toward thefront of the room, a blue fiberglass tub isperched on a hulking metal frame—an earlyversion of the bobsled simulator. Nestlednext to it, looking something like a home-made respirator, stands a one-of-a-kindpitching machine that can launch baseballsat 240 kilometers per hour spinning in anydirection. A dent in the wall attests to thepunch that such supercurveballs pack.

Hubbard has toys that any weekend war-rior would envy, but he is far from a natural-born jock. “I love sports,” he says, but “Iwas never a very successful athlete. I wasalways a scrawny kid.” Growing up in Alta-vista, a small town in central Virginia, Hub-bard played baseball fanatically but neverblossomed into a star. He didn’t really devel-op physically until he entered the U.S. Mili-tary Academy at West Point, New York,from which he graduated in 1964. Later inlife, Hubbard played squash avidly until hissecond back surgery sidelined him for good.

After a stint in the Air Force, Hubbardearned his doctorate in 1975 from StanfordUniversity in California, doing his thesisresearch on control systems of automobileengines. For the first several years of hiscareer, Hubbard stuck to conventional en-

gineering. Then, during a sabbatical atthe University of Cambridge in 1981,he became fascinated with the mechan-ics of walking. He soon began applyinghis engineering skills to other types ofhuman motion, especially sports. “Itwould have been an awfully boring 30years if I’d just been doing the same oldstuff,” he says.

Following his heart has meant aban-doning traditional funding sources. “Al-most nobody pays us to do this,” he says.Hubbard and his students make do pri-marily with four-figure grants from theU.S. Olympic Committee and varioussports federations. So they look for waysto learn more by spending less. For ex-ample, in one corner of the lab a tennisball half-covered with aluminum foilhangs from two pieces of string. AlbertJordan, an undergraduate, sets the ballspinning and swinging in front of a radargun. He hopes that radar waves reflectedby the half-silvered ball will reveal theball’s spin—a first step toward trackingthe spin of a batted baseball. It’s also,Jordan says, “a way to do an experimentwithout spending any money.”

Despite the relatively spartan condi-


CR

ED

IT:D

EB

BIE

ALD

RID

GE/U

C D

AV

IST E S T I N G H U M A N L I M I T S S

PE

CIA

LS

EC

TIO

N

N E W S

Engineering Peak PerformanceMechanical engineer Mont Hubbard can tell athletes how to do it faster,higher, and farther. But will they listen?

Having a ball. Engineer Mont Hubbard applies his exper-

tise in mechanics to sports “because it’s fun.”


tions, Hubbard has no troubleattracting students. Master’sdegree student John Kockel-man once owned a bungee-jumping company and per-formed stunts for televisioncommercials; he’s studyingthe properties of bungeecords. Doctoral student Alison Sheets competed ingymnastics from childhoodthrough college; she’s study-ing the dynamics of thewomen’s uneven bars. “Mybrain is far better at gymnas-tics than my body ever was,”Sheets says. “I can still do theskills, and sometimes I feellike I’m doing them betterbecause I definitely under-stand them a lot better now.”

Game planAlthough Hubbard has eclectic tastes inresearch problems, he approaches them allwith a well-defined method. First, he triesto tease out the essential elements and fea-tures of an activity. For example, in ana-lyzing the flight of a ski jumper, Hubbardand colleagues model the jumper as an as-semblage of weighty links representingskis, legs, torso, and head. Using the math-ematical methods of mechanical engineer-ing, the researchers crank out differentialequations that describe the forces and mo-tions of the various parts, which they gen-erally solve with computers. They system-atically alter the variables that the athletemight control—such as the angle betweena ski jumper’s skis and the wind whippingpast—to determine optimal values. Final-ly, if all goes well, Hubbard’s team devel-

ops a methodology to help athletes reachpeak performance.

Hubbard has applied this approach tomyriad sports, most successfully to thethrowing events in track and field. In theshot put, discus, javelin, and hammer throw,a thrower can adjust only a handful of an-gles and rates of rotation. Given an athlete’sbasic strength and speed, an engineer cancalculate exactly how far the athlete willever be able to heave the object. “That’s avery beautiful concept to me,” Hubbardsays. “In a sense, you’re allowing people toachieve the very best they can, given theirphysical limitations.”

But such ruminations also underscorethe appeal of muscle-building anabolicsteroids, he says. Once an athlete has per-fected a technique, the only way to throwfarther is to boost strength and arm speed—for some, a pursuit that knows no bounds.

Athlete attitudesHow athletes feel aboutHubbard’s methods de-pends on whom you ask.His javelin-training systemprovided valuable informa-tion, says Donna Mayhew, aseven-time U.S. nationalchampion and an Olympianin 1988 and 1992. Thejavelin “is such a technicalsport, any feedback you canget really helps a lot,” shesays. “Little changes intechnique are going tomake big changes in dis-tance.” Hubbard’s analysisshowed that Mayhew tend-ed to throw the javelin withits nose pointed several de-grees too high, she says.

Bobsled driver Brian Shimer, a five-time Olympian who steered the UnitedStates to a bronze medal in 2002, says thatthe bobsled simulator is “great for off-season training and for working on hand-eye coordination.” But the simulator can-not reproduce the feel of a bobsled ride,especially the crushing accelerations in theturns. So riding it might actually dull a dri-ver’s well-honed touch, says Shimer, whonow coaches U.S. bobsled drivers. “Oncethe season starts,” he says, “I would notwant to get in a simulator.”

And in the tradition-steeped world ofbaseball, scientific analysis takes a back seatto good old-fashioned coaching. Hubbard’sstudy of hitting curveballs and fastballsmakes “interesting conversation,” says GaryMatthews, batting coach for the ChicagoCubs, but it won’t change his coachingstyle. “It’s more important to the scientiststhan it would be to me,” he says. “What’simportant to me is that a guy has a lot ofheart and can move the fastball.”

Still, Hubbard and other sports engineerssay they are slowly making inroads withcoaches and athletes—and their numbers,although small, are growing. “There’s a so-ciety of sports engineers that just didn’t ex-ist 10 years ago,” says Rod Cross, a physi-cist at the University of Sydney, Australia,who studies tennis. This September, Hub-bard and his colleagues at Davis will hostthe Fifth International Conference on theEngineering of Sport.

Hubbard says he doesn’t worry how—or if—athletes use his ideas. He’s too busymoving from project to project. “I don’twant to be teaching people how to throwthe javelin,” Hubbard says. “I want to bethinking about new stuff.” For engineers atthe top of their form, that’s the whole pointof the game. –ADRIAN CHO C

RED

IT:D

EB

BIE

ALD

RID

GE/U

C D

AV

IS


SP

EC

IAL

SE

CT

ION

Virtually flying. Hubbard and his team tune up a bobsled simulator.

Long Gone or Gone Wrong?

Last November, Mont Hubbard and colleagues argued in the American Journal of Physicsthat a well-hit curveball would sail farther than a perfectly struck fastball—even though

the curveball moves slower and packs less energy, both before and after it’s walloped.

That’s because a batted ball travels farther if it has more backspin to give it aerodynamic

lift. The top-spinning curveball approaches the batter already turning in a direction that

increases the backspin of the batted ball. On the other hand, the back-spinning fastball

comes at the batter spinning the wrong way, which decreases the backspin of the out-

going orb. As a result, an optimally struck curveball will travel around 455 feet (138 me-

ters), about 12 feet (3.5 meters) farther than a well-hit fastball.

Not so, contends Robert Adair, a physicist at Yale University and author of The Physicsof Baseball. Although he can’t say precisely what’s wrong with Hubbard’s calculations, he

claims that balls moving at the speeds Hubbard quotes just don’t go that far, so Hubbard

and colleagues must have overestimated the lifting effect of spin. For his part, Gary

Matthews, batting coach for the Chicago Cubs, says Hubbard’s team may be right—

especially if the curveball “hangs” high in the strike zone. “A hanging curveball will go a

long, long way,” he says. Although Matthews may not be an expert in aerodynamics, he

did belt 234 homers in 16 seasons in the major leagues. –A.C.

APOBEC-Mediated Editing ofViral RNA

Kate N. Bishop, Rebecca K. Holmes, Ann M. Sheehy, Michael H. Malim*

Viral sequence variation frequently plays anessential role in viral replication and patho-genesis. For instance, sequence changes canunderlie alterations in viral phenotype, es-cape from host immune responses, and theacquisition of drug resistance. Usually, suchchanges are due to the fixation of copyingerrors made during genome replication. Se-quence variation may also arise through thepost-synthesis editing of RNA or DNA: Mea-sles, hepatitis delta, and polyoma viral RNAs,for example, are presumed substrates for ad-enine deamination (1). Editing of human im-munodeficiency virus (HIV) DNA by the hu-man cytidine deaminase APOBEC3G (hA3G)has also received much recent scrutiny. HIVparticles carrying this enzyme yieldminus-strand reverse transcripts inwhich multiple cytosine (C) residuesare deaminated to uracil (U). The re-sulting hypermutation [guanine (G)to adenine (A) on plus strands] andcDNA instability combine to inhibitviral infectivity. Normally, the viralprotein Vif protects HIV from hA3Gby inducing its degradation and ex-clusion from virions (2).

Beyond hA3G, mammalian genomesharbor a variety of additional RNA/DNAcytidine deaminases (3). The foundermember of this APOBEC family isAPOBEC1 (human A1, or hA1), the cat-alytic subunit of a complex that deami-nates C6666 in apolipoprotein B mRNAin human gastrointestinal tissue to yielda premature stop codon. To determinewhether other APOBEC proteins canmutate HIV and/or regulate infectivity,wild-type or vif-deficient (‚vif) viruseswere produced in the presence of variousfamily members, and viral infectivitieswere measured (Fig. 1A) (4, 5). Asexpected, hA3G profoundly inhibitedthe ‚vif virus but had a mild effect onits wild-type counterpart. Surprisingly,whereas hA1 had negligible effects, ratA1 (rA1) was a potent suppressor ofboth viruses. Although the basis for thedichotomy between rA1 and hA1(which share �70% amino acid se-quence identity) is not understood, wespeculate that either hA1 is intrin-sically incapable of influencing HIVinfection or that 293T cells lack

cellular cofactors critical for this activity (6).Because hA3G induces excessive C-to-U

deamination of minus-strand HIV cDNA, weasked whether this extended to rA1. ViralcDNAs were recovered from infected cells, anda nef/3�–long-terminal-repeat (LTR) fragmentwas then amplified by polymerase chain reaction(PCR), cloned, and individual sequences deter-mined (Fig. 1B). The mutations induced byhA3G were �95% G-to-A transitions (5). How-ever, this was only the predominant mutation(�70%) induced by rA1, with many C-to-Tmutations also being evident (�22%). Theseparticular mutations could have arisen throughdeamination of unpaired plus-stranded cDNA orvirion RNA.

To address the latter possibility, cell-free viri-ons were purified and a nef/3�-LTR region of thegenomic RNA was converted to cDNA in vitro,amplified by PCR, cloned, and sequenced (Fig.1C). There was a clear accumulation of C-to-T(U in the template RNA) changes in the RNAsof HIV produced in the presence of rA1 (17events in 23,982 bases, a frequency close to thatseen with DNA sequencing). Thus, HIV RNA aswell as DNA can be a substrate for APOBEC-mediated deamination, making it plausible thatC-to-U editing of either could contribute to thesequence diversity that marks natural HIV infec-tion (7). In both cases, this will require expres-sion of APOBEC proteins and their presumedcofactors in the natural targets of HIV infection,namely T lymphocytes and macrophages.

These observations allow speculation (i)that APOBEC-mediated C-to-U editing maycontribute to the sequence variation of virus-es that replicate entirely through RNA, ofwhich there are many; and (ii) that additionalcellular RNA substrates might exist for theAPOBEC enzymes. Both possibilities willrequire further rigorous experimental investi-gation. It has also been demonstrated thathA3G inhibits the association of hepatitis Bvirus RNA with viral cores—in this case, inthe absence of viral RNA/DNA deamination(8). Thus, APOBECs have the potential toaffect their substrates through at least threemechanisms: DNA editing, RNA editing, anda nonediting pathway. The possibilities forAPOBEC involvement in the regulation anddysregulation of cell function (6, 7), as wellas in innate resistance to invading nucleicacids, may therefore be considerable.

References and Notes1. B. L. Bass, Annu. Rev. Biochem. 71, 817 (2002).2. X. Yu et al., Science 302, 1056 (2003).3. A. Jarmuz et al., Genomics 79, 285 (2002).4. Materials and methods are available as supporting

material on Science Online.5. K. N. Bishop et al., Carr. Biol., in press; published online 24

June 2004 (10.1016/50960982204004683).6. S. Anant, N. O. Davidson, Trends Mol. Med. 9, 147 (2003).7. R. C. Beale et al., J. Mol. Biol. 337, 585 (2004).8. P. Turelli, B. Mangeat, S. Jost, S. Vianin, D. Trono,Science 303, 1829 (2004).

9. Supported by the U.K. Medical Research Council, theBiotechnology and Biological Sciences ResearchCouncil, the Royal Society (A.M.S.), and the ElizabethGlaser Pediatric AIDS Foundation (M.H.M.).

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/645/DC1Materials and MethodsReferences and Notes

24 May 2004; accepted 18 June 2004

Department of Infectious Diseases, Guy’s, King’s andSt. Thomas’ School of Medicine, King’s College Lon-don, London, SE1 9RT, UK.

*To whom correspondence should be addressed. E-mail: [email protected]

120

100

80

60

40

20

0

% In

fect

ivity

WT

hA3GA C G T

ACG 25 1 1T

A C G TA 1C 1G 105 2T 1

From

To

n=6500bp n=6500bp

rA1A C G T

A 4 5 1C 1 2 43G 93T

A C G TA 2 1C 1 1 13G 83 1 1T 2

n=35100bp n=25350bp

Wild Type ∆Vif

*

* * * * ** * ** * *

** * * * *

hA3G/WT

hA3G/∆Vif

rA1/WT

rA1/∆Vif

A

B

C

No APOBEC

hA3GhA1

rA1

From

From

From

To

To To

∆Vif

Fig. 1. (A) Infectivity of HIV produced in the presenceof hA3G, hA1, or rA1; error bars show standard devia-tion. WT, wild type. (B) Plus-strand cDNA mutationspectra for the indicated virus/APOBEC pairs; n � thetotal number of sequenced base pairs. (C) rA1 deami-nates viral RNA. Clones (n� 21) were analyzed for eachvirus, with all changes marked onto a single horizontalline representing the entire amplicon; C-to-T changesare indicated by asterisks.

BREVIA


Engineered Interface ofMagnetic Oxides

Hiroyuki Yamada,1,2 Yoshihiro Ogawa,3 Yuji Ishii,1,2

Hiroshi Sato,1,2 Masashi Kawasaki,1,4 Hiroshi Akoh,1,2

Yoshinori Tokura1,3,5

Interface-selective probing of magnetism is a key issue for the design andrealization of spin-electronic junction devices. Here, magnetization-inducedsecond-harmonic generation was used to probe the local magnetic propertiesat the interface of the perovskite ferromagnet La0.6Sr0.4MnO3 with nonmag-netic insulating layers, as used in spin-tunnel junctions. We show that bygrading the doping profile on an atomic scale at the interface, robust ferro-magnetism can be realized around room temperature. The results should leadto improvements in the performance of spin-tunnel junctions.

In the development of semiconductor technol-ogy, a major role was played by the elucidationof electronic states at heterointerfaces such aspositive-negative (p-n) junctions and metal/semiconductor interfaces. In the p-n junction,for example, a depletion layer and built-in elec-tric field are formed at the interface so as tounify the Fermi level all over the junc-tion. Analogously, transition-metal-oxide com-pounds, as emergent electronics materials withcorrelated electrons, often exhibit nontrivialand potentially useful electronic properties atheterointerfaces. In most cases, however, suchinterface properties cannot be explained in

terms of conventional band pictures (1).Moreover, the charge state in the stronglycorrelated electron oxides is closely cou-pled with the spin and orbital degrees offreedom. Therefore, magnetic oxide inter-faces provide a challenging arena to ex-plore new multifunctional materials (2).

Interface magnetism is also important for thedevelopment of devices with strongly correlatedelectron oxides. For example, the perfectly spin-polarized ferromagnet La1-xSrxMnO3 (LSMO)could be one of the best candidates for magneticrandom-access memory if we could fully use itspotential in spin-tunnel junctions. However, inthe actual junctions—for example, LSMO/SrTiO3 (STO)/LSMO—the tunnel magnetore-sistance (TMR) has been much smaller than thatexpected from the half-metallic nature of LSMO,and the response has been diminished at temper-atures far below Tc (3, 4). The possible origin ofthe decreased MR is severe deterioration of theferromagnetism locally occurring near theLSMO/STO interfacial region (5–7). Therefore,selective and quantitative evaluation of interfa-

cial spin state may provide a key to realizinghigh-performance magnetic devices.

Nonlinear magneto-optical effects can probethe interface magnetism (8, 9), as has recentlybeen demonstrated for oxide “tricolor” superlat-tices, where ferromagnetic LSMO layers andinsulating STO and LaAlO3 (LAO) layers wereaccumulated one by one in an ABCABC. . .structure. The STO/LSMO/LAO superlattice ex-hibits very large magnetization-induced second-harmonic generation (MSHG) and a resul-tant nonlinear magneto-optical Kerr effect(NOMOKE) (10, 11). This “tricolor” spin super-lattice is regarded as an artificially constructednoncentrosymmetric ferromagnet, where sym-metries of space inversion and time reversal areboth broken simultaneously. The generic fea-ture of such a noncentrosymmetric ferromagnetshould be extended to various heterointerfaces ofstrongly correlated electron oxides.

We show that such an interface magne-tism can indeed be probed by MSHG fromsingle heterojunctions between LSMO andSTO or LAO, and we have found a robustferromagnetic interface of LSMO. It can beobtained by modulating the interface dopingprofile on an atomic scale, which may openup a new path to LSMO-based spin-injectiondevices such as TMR junctions operating atroom temperature.

The schematic side view of the hetero-structures used in the present study is depict-ed in Fig. 1A. For detecting the interfacemagnetization at a single magnetic interface,we prepared bilayer films composed of aninsulating cap layer (STO or LAO) with athickness of 5 unit cells (u.c.) (i.e., 2 nm) anda magnetic LSMO bottom-layer grown onSTO(001) substrates (Fig. 1A, left). TheLSMO layer is as thick as 300 u.c (120 nm).Because the optical absorption coefficients inLSMO for incident beam (� � 800 nm) andSH light (� � 400 nm) are about 1 � 105

cm�1, the contribution from the bottom film/

1Correlated Electron Research Center (CERC), Nation-al Institute of Advanced Science and Technology(AIST), Tsukuba 305-8562, Japan. 2Core Research forEvolutional Science and Technology (CREST), JapanScience and Technology Agency (JST), Tsukuba 305-8562, Japan. 3Spin Superstructure Project (SSS),Exploratory Research for Advanced Technology(ERATO), Tsukuba 305-8562, Japan. 4Institute for Ma-terials Research, Tohoku University, Sendai 980-8577,Japan. 5Department of Applied Physics, University ofTokyo, Tokyo 113-8656, Japan.

Fig. 1. (A) Atomicstacking sequencesfor perovskite-typemagnetic oxide inter-faces. (Left) LSMOfilm capped with a5-u.c. (2-nm) layer ofLaAlO3 (LAO) orSrTiO3 (STO). (Right)A 2-u.c. layer of un-doped LMO is insert-ed between STO andLSMO layers to grade

the hole concentration in the manganite layer near the interface. In these films,the top surface is terminated by the MnO2 or BO2 (B � Ti or Al) layer (18). Thearrows depict possible arrangements of magnetic moments at the respectiveMnO2 layers, showing considerable spin canting near the interface (left) androbust ferromagnetism (right). (B) (Left) RHEED oscillation during the growth ofthe STO/LMO/LSMO heterostruture. (Right) Atomic force microscopy image forthe top surface of STO/LMO/LSMO. Atomically flat terraces with the single-u.c.steps (0.4 nm high) are clearly seen.

REPORTS


substrate interface is negligibly small (�1%).All the films were fabricated by pulsed laserdeposition (PLD) in a layer-by-layer growthmode while observations were made of theintensity oscillation of the specular spot inreflection high-energy electron diffraction(RHEED) (Fig. 1B). The resultant film sur-faces are atomically flat with 200-nm-wideterraces and 0.4-nm-high steps. We also notethat there should be no detectable interfaceroughness or disorder, judging from the tran-sition electron microscopy image and x-raydiffraction for the “tricolor” superlatticegrown in a similar way (12). Therefore, wemay conclude that exact control of interfaceatomic stacking can be realized. Four-circlex-ray diffraction measurements indicate thatthe in-plane lattice constants are identicalwith that of the substrate (STO), ensuringcoherent epitaxy throughout the bilayer film.Magnetization of the bilayer film (Mfilm) wasmeasured with a superconducting quantuminterference device (SQUID) magnetometer,indicating in-plane magnetic anisotropy,well-defined Tc of 340 K, and saturated mag-netization of 3.6 �B / Mn for all the samples.

Figure 2A illustrates the experimentalconfiguration for the MSHG measurements(13). The external or built-in fields inducemagnetization (M), electric polarization (P),and toroidal moment (T � P � M), whichgives MSHG (Fig. 2A, bottom). The SH lightintensity (I), measured in a magnetic field (H)of � 0.05 T, was plotted against the analyzerangle (), as shown in Fig. 2C. MSHG cor-

responds to the s(y)-polarized component( � 90°) in the SH light. A phase shift in theI- curves for � H gives twice the NOMOKEangle, from which we can deduce the inten-sity of MSHG. We define the square root ofthe MSH intensity as the interface magneti-zation (Mint), which is plotted against T inFig. 2B. In the upper panel, the Mint-T plot isnormalized with Mint(50 K) and comparedwith the Mfilm-T curve.

The MSHG data shown in Fig. 2C indi-cate a large variation of magnetic propertiesbetween STO/LSMO and LAO/LSMO inter-faces. The STO/LSMO interface hardlyshows MSHG or NOMOKE even at 50 K,which directly demonstrates the existence ofa dead layer at the interface and explains theinferior TMR response found in LSMO/STO/LSMO junctions (5). At the LAO/LSMO in-terface, by contrast, large MSHG is observed,and the ferromagnetism survives even at 250K, although the onset T of the MSHG signalas the interface Tc is not so high or welldefined as the Tc of the bulky thick films. Thepresent observation is consistent with thefinding of the studies on transport propertiesof the related superlattices that the LAO andSTO form robust and susceptible interfaceswith LSMO, respectively (10, 11).

Figure 3 exhibits the magnetic field depen-dence of Mint as deduced by the square root ofMSHG intensity, measured at T � 50 K. This“interface M-H curve” enables us to decomposeMint into the spontaneous magnetization andfield-induced part. To clearly distinguish the

two components, Mint is fitted with the sponta-neous magnetization plus the H-linear functionand normalized with Mint(7 T) (Fig. 3, top). Themagnetic field of 7 T increases the spontaneousmagnetization by 200% and 70% for STO/LSMO and LAO/LSMO interfaces, respective-ly. This means that the antiferromagnetic spincanting occurs at the STO/LSMO interface,while the ferromagnetic spin arrangement ismuch less canted by LAO.

In bulk crystals, LSMO (x � 0.4) has thehighest Tc of 370 K among the La1-xSrxMnO3

series (14), but the LSMO (x � 0.4) film grownon STO is on the verge of an A-type (layered)antiferromagnet with reduced Tc(340 K) as aresult of the tensile strain (15). This instabilitytoward the antiferromagnetic phase explainswhy the ferromagnetic spin ordering in LSMO(x � 0.4) is so easily reduced at the interface.Moreover, STO produces a dead layer in theadjacent LSMO more aggressively than doesLAO, because the STO layer works as ahole-donating layer. It has been considered thatthe valence-mismatched interface composedof the stacking sequence -TiO2-SrO-MnO2-La0.6Sr0.4O- induces the charge transfer andthat the overdoped LSMO is dominated by theantiferromagnetic spin canting (5–7, 10, 16).Therefore, the underdoped LSMO (preferablyx � 0.3) layer may enable us not only tostabilize the ferromagnetism but also to com-pensate the charge transfer from STO at theinterface. However, both high Tc and conduc-tivity are lost by decreasing x, where we con-front a dilemma.

A B C

y

xz

Analyzer angle

Cap layer

T

P

M

2

La Sr MnO0.6 0.4 3

H

0.10

0.05

0.00

Sqr

t MS

H In

tens

ity (

a. u

.)

300200100Temperature (K)

1.0

0.5

0.0

Nor

mal

ized

M

4

2

0

Magnetization (µ

B /Mn)

STO/LMO

LAOSTO

Mfilm

Mint

0.5

0.00.5

0.00.5

0.0180120600

Analyzer Angle (deg.)

50K

150K

250K

STO0.1

0.00.1

0.0

SH

Inte

nsity

(a.

u.)

0.1

0.0180120600

50K

150K

250K

LAO

+H -H

0.2

0.00.2

0.00.2

0.0180120600

50K

150K

250K

STO/LMO

ωω

θ

Fig. 2. (A) (Top) Schematic of experimental configuration for the mea-surement of MSHG. Polarization of incident light and applied magneticfield (H) are along y and x, respectively. The reflected SH light undergoesKerr rotation. (Bottom) Configuration of various electric and magneticmoments. P (//z) stands for electric polarization originating from non-centrosymmetric atomic stacking along the z direction. M (//x) showsmagnetization of LSMO along the film plane induced by a magnetic fieldH (//x). T (//y) indicates a toroidal moment, which is generated as T � P� M. (B) (Bottom) Temperature (T ) dependence of magnetization at theinterface (Mint), defined as the square root of MSHG intensity (s-in s-out

SH light). The applied magnetic field is 0.05 T. Red, blue, and greensymbols indicate the data for STO/LSMO, LAO/LSMO, and STO/LMO/LSMO interfaces, respectively. (Top) Mint(T) normalized with Mint(50 K),and T dependence of magnetization of the films (Mfilm) measured at 0.05T with a SQUID magnetometer (a black line). (C) SH intensities plottedagainst analyzer angle () at 50 K, 150 K, and 250 K for H and –H. denotes 0° and 90° for p- and s-polarized SH light, respectively. The dependence can be fitted to the formula �Pzcos � Pysin�2. Phase shiftinduced by the H reversal corresponds to twice the NOMOKE angle,which gives the magnitude of Mint.

R E P O R T S


To enhance the interface magnetizationwithout sacrificing the ferromagnetic and me-tallic characters in the film, we propose thecompositionally graded LSMO interfaces,where the bulk LSMO electrode has the optimaldoping level of x � 0.4, and x decreases grad-ually to 0 toward the insulating layer. This ideabears some analogy to the procedure to improvethe superconductivity grain boundary junction(17), in which partial Ca substitution on Y sitesin YBa2Cu3O7-x enhances the critical supercur-rent density (Jc), but at the same time suppress-es Tc. In our study, the doping profile wascontrolled on an atomic scale. The 2-u.c. layerof LaMnO3 (LMO) was inserted as a locallyunderdoped layer between STO (5 u.c.) andLSMO (x � 0.4; 300 u.c.) layers (Fig. 1A,right). We anticipated that the LMO layer,which is originally an antiferromagnet withTN � 140 K, should be hole-donated byLSMO(0.4) and STO layers and hence cancompensate for the interface effects. This com-positionally modulated interface shows largeMSHG and NOMOKE, which exceed by farthe STO/LSMO direct interface and are com-parable to those of the LAO/LSMO interface(Fig. 2B). In accord with this observation, thespin-tunnel junction equipped with such anatomically engineered interface, [LSMO(0.4)/LMO (2 u.c.)]/STO (2 nm)/[LMO (2 u.c.)/LSMO(0.4)], shows an improved (comparable)

device performance as compared with thedirect-interface junction, LSMO/STO/LSMO(LSMO/LAO/LSMO) (13). As shown in Fig. 3,Mint for this STO/LMO/LSMO interface is al-most H-independent above 0.1 T, thus confirm-ing that the nearly bulk-like ferromagnetism ismaintained even in the vicinity of STO. Thenormalized Mint(T) value of STO/LMO/LSMOreaches 0.5 at 250 K, which is close to a value(0.7) for normalized magnetization of bulk filmat the same temperature and is much larger thanthe value (�0.2) for the direct interface ofSTO/LSMO (x � 0.4).

References and Notes1. A. Ohtomo, D. A. Muller, J. L. Grazul, H. Y. Hwang,

Nature 419, 378 (2002).2. K. Ueda, H. Tabata, T. Kawai, Science 280, 1064

(1998).3. J. Z. Sun et al., Appl. Phys. Lett. 69, 3266 (1996).4. M. Bowen et al., Appl. Phys. Lett. 82, 233 (2003).5. Y. Ogimoto et al., Jpn. J. Appl. Phys. 42, L369 (2003).6. M. Izumi et al., Phys. Rev. B64, 064429 (2001).

7. M. Izumi, Y. Ogimoto, T. Manako, M. Kawasaki, Y.Tokura, J. Phys. Soc. Jpn. 71, 2621 (2002).

8. J. Reif, J. C. Zink, C.-M. Schneider, J. Kirschner, Phys.Rev. Lett. 67, 2878 (1991).

9. G. Spierings et al., J. Magn. Magn. Mater. 121, 109(1993).

10. H. Yamada, M. Kawasaki, Y. Ogawa, Y. Tokura, Appl.Phys. Lett. 81, 4793 (2002).

11. Y. Ogawa et al., Phys. Rev. Lett. 90, 217403 (2003).12. K. Kimoto et al., Appl. Phys. Lett. 84, 5374 (2004).13. Materials and methods are available on Science On-

line.14. A. Urushibara et al., Phys. Rev. B51, 14103 (1995).15. Y. Konishi et al., J. Phys. Soc. Jpn. 68, 3790 (1999).16. F. Pailloux et al., Phys. Rev. B66, 014417 (2002).17. G. Hammerl et al., Nature 407, 162 (2000).18. M. Izumi et al., Appl. Phys. Lett. 73, 2497 (1998).19. We thank T. Arima, J. Matsuno, and A. Sawa for

enlightening discussions.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/646/DC1Materials and MethodsFig. S1References and Notes

7 April 2004; accepted 29 June 2004

Real-Space Observation ofMolecular Motion Induced byFemtosecond Laser Pulses

Ludwig Bartels,1 Feng Wang,2 Dietmar Moller,2 Ernst Knoesel,3

Tony F. Heinz2*

Femtosecond laser irradiation is used to excite adsorbed CO molecules on aCu(110) surface; the ensuing motion of individual molecules across the surfaceis characterized on a site-to-site basis by in situ scanning tunneling microscopy.Adsorbate motion both along and perpendicular to the rows of the Cu(110)surface occurs readily, in marked contrast to the behavior seen for equilibriumdiffusion processes. The experimental findings for the probability and directionof the molecular motion can be understood as a manifestation of strongcoupling between the adsorbates’ lateral degrees of freedom and the substrateelectronic excitation produced by the femtosecond laser radiation.

Scanning tunneling microscopy (STM),through its direct imaging on the atomiclength scale, has dramatically advanced ourability to probe the geometric and electronicstructure of surfaces. The introduction ofvariable temperature STM has further en-abled researchers to follow atomic motion onsurfaces in real time. Although this capabilitypermits the examination of processes on themillisecond time scale, the inherent charac-teristics of STM imply that new approachesare required to reach the intrinsic time scaleof atomic motion, which is in the pico- tofemtosecond range.

Several researchers have explored gatingand correlation techniques to improve thetime resolution of STM measurements (1–3),but their focus has been on electron dynamicsrather than on the nuclear motion that corre-sponds to chemical processes. In a separateline of research, the utility of femtosecondlaser radiation to induce surface chemicalreactions and to clarify rates of energytransfer and nuclear motion has been dem-onstrated (4–14 ). Although they access theultrafast dynamics of surface reactions,these methods do not provide the directreal-space imaging of STM.

We describe here the successful combina-tion of the atomic-scale spatial resolutionprovided by STM with ultrafast surface dy-namics driven by femtosecond laser excita-tion. In this fashion, we are able to determinethe initial and final configurations of individ-ual molecules undergoing nonequilibriumsurface diffusion induced by electronic exci-

1Pierce Hall, University of California, Riverside, CA92521, USA. 2Department of Physics and Department ofElectrical Engineering, Columbia University, New York,NY 10027, USA. 3Department of Physics and Astrono-my, Rowan University, Glassboro, NJ 08028, USA.


1.5

1.0

0.5

0.0

Nor

mal

ized

M

0.3

0.2

0.1

0.0

Sqr

t MS

H In

tens

ity (

a. u

.)

6420Magnetic Field (T)

4

2

0

Magnetization (µ

B /Mn)

STO/LMO

LAO

STO

Mfilm

Mint

Fig. 3. Magnetic field (H) dependence of thesquare root of MSHG intensity, representingthe interface magnetism (Mint), measured at 50K. Mint (bottom) and normalized Mint [Mint(H)/Mint(7 T)] (top). STO/LSMO, LAO/LSMO, STO/LMO/LSMO interfaces are displayed with red,blue, and green symbols, respectively. The solidlines are the fit, assuming spontaneous magne-tization (intersection to the vertical axis) and Hlinear increase inMint from the canted states byan external field. A black line in the top panelshows the H dependence of Mfilm at 5 K.

R E P O R T S


tation. Given the critical role that surfacediffusion plays in phenomena as varied ascrystal growth and catalytic activity of sur-faces, many techniques have been applied toprobe the diffusion of adsorbed atoms andmolecules under equilibrium conditions (15,16). By exciting the molecules electronicallyrather than thermally, we can now examinesurface diffusion processes occurring undernonequilibrium conditions on the ultrafasttime scale. The measurements, carried out onthe CO/Cu(110) system, reveal adsorbatemotion with different propensities for dis-placement along and perpendicular to theatomic rows of the surface compared withthose previously observed under conditionsof thermal equilibrium. The experimental re-sults demonstrate the importance of the elec-tronic adsorbate-substrate coupling for sur-face migration, as well as the existence ofdistinctive new surface phenomena on theultrafast time scale.

We chose the CO/Cu(110) system for studybecause its equilibrium and dynamical propertieshave been well characterized. CO adsorbs in amolecular state at on-top sites of the Cu(110)surface and is stable against thermal desorptionup to a temperature of �190 K (17). With theuse of STM imaging, Briner et al. (18) foundthat thermally induced diffusion starts at temper-atures around 40 K and proceeds exclusivelyalong the close-packed �110� rows of the sub-strate. Researchers have also recently investigat-ed STM-induced lateral motion of adsorbed CO(19–23). These measurements in the spatial do-main have been complemented by experimentalinvestigations of femtosecond laser-induced pro-cesses, including research on desorption (5, 7)and adsorbate-substrate vibrational coupling(24–27), as well as several related theoreticalstudies (28–31). These experimental and theoret-ical investigations have revealed the importanceof nonadiabatic coupling of the adsorbate to thehigh electronic temperature of the substrate pro-duced by femtosecond laser excitation.

During the experiment (32), the Cu(110)sample was held at a temperature of 22 K,well below the threshold for thermally drivensurface diffusion of CO. The CO coverage of�0.01 monolayer permitted isolated mole-cules to be studied. STM images were takenat tunneling currents of �100 pA with a biasof a few hundred millivolts. Under theseconditions, no STM-induced changes of theCO-covered surface were observed. Laser ra-diation was provided by a regeneratively am-plified mode-locked Ti:sapphire laser. Weused the second harmonic of this source (405nm) to irradiate the sample with 200-fslaser pulses at a repetition rate of �1 kHz.

We recorded the positions of CO mole-cules on the Cu(110) surface by STM beforeand after irradiation by the laser and therebytraced the motion of individual adsorbed mol-ecules. Although conceptually simple, thisprocedure requires careful consideration fromthe experimental standpoint. The principalcomplication is that the surface cannot beirradiated with the STM tip in its tunnelingposition because of thermal expansion.

The experimental approach adopted to cir-cumvent this problem is sketched in Fig. 1. Afteracquisition of an initial STM image, the tip isquickly withdrawn from the substrate by �1 �mwith the vertical piezoelectric drive. With the tipretracted, the surface is exposed to the fs-pulselaser radiation. The STM tip is then allowed toreapproach the surface. To use the STM to fol-low atomic-scale motion, we must be able toimage precisely the same region of the surface.Control experiments, as indicated in Fig. 2, Aand B, confirm this capability and verify that themolecules retain exactly the same positions. Incontrast, Fig. 2, B and C, show STM images ofadsorbed CO molecules before and after expo-sure to fs-pulse laser radiation. Because the prob-ability of laser-induced events is low for a singlepulse, we typically applied 1000 or more pulsesbetween successive images. In our measure-ments, a total of �3000 isolated molecules were

analyzed, half of them subject to laser irradiationand half in control experiments.

Under irradiation by the femtosecond la-ser pulses, we observed surface diffusion assummarized in Fig. 3. In contrast to the ther-mal behavior, adsorbate migration occursacross the close-packed rows of the substrate(in the �001� direction) as well as alongthem (in the �110� direction). Also inmarked distinction to conventional thermalbehavior, desorption occurs at a rate similarto surface diffusion.

An important experimental issue concernsthe determination of the absorbed laser fluence,which has been shown to be a critical experi-mental parameter (4–13). The geometry of thepresent experiment makes a direct determina-tion of the absorbed fluence difficult: Even withthe STM tip retracted, we expect that its pres-ence will alter the strength of the laser field atthe surface. We can, however, apply an internalcalibration scheme on the basis of the presenceof laser-induced CO desorption. To convert theobserved desorption rates into an absorbed flu-ence, we apply the phenomenological modeldeveloped by Struck et al. (7) for Cu(001), withan adjustment to reflect the empirical Arrheniusrate law for CO desorption from Cu(110). Giv-en the sharp dependence of the desorption rateon the absorbed laser fluence, we believe thatthis procedure yields a relatively precise localvalue of this parameter of 30 � 3 J/m2.

We now discuss the mechanism for fem-tosecond laser-induced diffusion and the ori-gin of the anomalous branching ratio for mo-tion in the �110� and �001� directions.The absorbed laser radiation produces elec-tron-hole pairs that thermalize rapidly to aFermi-Dirac distribution. However, this elec-tronic temperature can far exceed that of thephonons (25). To model this transient sub-strate excitation, we have solved coupled dif-fusion equations for the substrate electronicand lattice temperatures at the surface (Fig. 4). Atransient electronic temperature approaching

Fig. 1. Overview of experi-mental procedure. (A) Initially,a molecularly resolved STMimage of the surface is ac-quired. (B) The tip is retractedto the limit of the feedbackloop, the vertical piezo isswitched to a high-voltagesupply, and the tip is retractedmore than 1 �m from thesample. Laser radiation is thenapplied. (C) The tip is broughtback toward the surface, andthe feedback loop becomesoperational. After creep com-pensation, the surface is reap-proached and stable tunnelingis achieved. Finally, we restartthe STM imaging. In the fol-lowing control cycle, the en-tire procedure is repeated without laser irradiation.

R E P O R T S


3000 K is achieved for a time of �1 ps; heatingof the lattice also occurs, but the lattice onlyreaches a temperature of 140 K in a trans-

ient of a few tens of picoseconds in duration.With this description of the substrate ex-

citation, we first consider whether the ob-

served diffusion arises from a transient ther-mal process associated with the (relatively)slow equilibrium heating in the metal. Withthe use of data from Briner et al. (18), we findthat the expected number of hops along theatomic rows falls short by more than an orderof magnitude. Also, CO hopping across therows is negligible under conventional thermalconditions and should, contrary to our exper-imental findings, still be relatively minor fortransient thermal excitation. Further, a purelythermal mechanism could not explain theconcurrent CO desorption process observedunder laser excitation.

These considerations strongly suggest thatan electronically driven process is responsi-ble for the observed surface diffusion. Al-though theoretical attention has focusedlargely on the role of electronic temperaturein driving desorption processes, the notion ofelectronic coupling to coordinates relevantfor lateral adsorbate motion is also inherent-ly reasonable, because strong adsorbate-substrate charge transfer is expected as theadsorbate moves from one surface bindingsite to the next. Indeed, in the calculations ofKindt et al. (30), it was explicitly noted thatelectronic coupling to modes other than themolecule-surface vibration for CO/Cu(100)should play an important role.

Here, we introduce a simple phenome-nological model of such an electronicallydriven diffusion process. With the use ofonly parameters from the literature, we canroughly reproduce the experimental rate fordiffusion along the atomic rows of theCu(110) surface and account for the anom-alous behavior for diffusion across theserows. Our starting point is the experimentalArrhenius rate law reported by Briner et al.(18) for CO diffusion along the �110�direction. We now, however, consider en-ergy transfer separately from the electronicand lattice degrees of freedom of the sub-strate, following a model previously intro-duced to examine coupling between thesubstrate and adsorbate vibrations:

dUad/dt � (Uel – Uad)/�el � (Uph – Uad)/�ph

(1)

where Ux � hv/exp(hv/kTx – 1) denotes theenergy content of a harmonic oscillator corre-sponding to the frustrated translational mode( � 1.1 1012 s�1) at the specified temper-ature Tx (where ad, el, and ph indicate theadsorbate, electronic, and phonon degrees offreedom, respectively) and h and k are the usualPlanck and Boltzmann constants. The couplingparameters (�el � 5.1 ps and �ph � 4.2 ps) forfrustrated translational motion, which is as-sumed to be the reaction coordinate for surfacediffusion, were obtained experimentally byGermer et al. for the CO/Cu(001) system(24, 25 ). Although we have included for

Fig. 2. Portion of STM images (1.00 V, 200pA, and 50 Å by 25 Å) of CO molecules(dark regions indicated by circles) on theCu(110) surface taken at high scan speed.The rows of surface atoms along the �110�direction are highlighted on the basis of sep-arate slow scans. The tip was removed andreapproached without optical excitation be-tween images (A) and (B). Although a slightdrift is seen, precisely the same region of thesurface has been found. Between images (B)and (C), fs-pulse laser excitation was applied.(C) has been numerically corrected for drift.One adsorbed CO molecule (with an initialposition indicated by the dashed yellow cir-cle) moved to a new position (red circle) onesite along the �110� direction. The brightspot in the bottom of the images is a metalcluster that was deposited by controlledtip-sample contact for tip preparationand as a registry marker. Analysis of hun-dreds of such images led to the data setpresented in this work.

Fig. 3. (A) Schematic representa-tion of Cu(110) indicating thecrystallographic axis and inter-atomic distances. (B) Summaryof experimental results of fslaser-induced diffusion and de-sorption for CO-Cu(110) mea-sured with a total of �3000adsorbed molecules. The ratio ofhopping across the close-packedrows (�001�) to hopping alongthe rows (�110�) is 0.34.

Fig. 4. (A) Calculated temperature pro-files for the substrate electrons (Tel) andphonons (Tph), and the adsorbate frustrat-ed translational mode (Tad) under laserirradiation of the Cu(110) sample. A peakelectronic temperature Tel,max � 3000 Kis reached, in contrast to a peak phonontemperature Tph,max � 140 K. (B) TheCO diffusion probability across theclose-packed rows (�001�) and alongthe rows (�110�) from the calculatedincrease in the adsorbate temperature.

R E P O R T S


completeness both electronic and phononiccouplings, the observed surface hopping isdriven by the high electronic temperature.

The predicted temperature of the trans-lational degree of freedom of the adsorbedCO molecules is shown in Fig. 4. We cal-culate a hopping probability per laser pulseof 2 10�6. Given the severe approxima-tions in this treatment, the agreement withthe experimental result of 9 10�6

is satisfactory.Can this analysis also explain the anoma-

lous behavior with respect to adsorbate mo-tion perpendicular to the atomic rows? Here,we cannot rely on an established Arrheniusrate law; all we know is that measurablehopping perpendicular to the atomic rows hasnot been observed under conventional ther-mal conditions. For purposes of a semiquan-titative discussion, we assume that the lack ofthermal diffusion in the perpendicular direc-tion is attributable primarily to a larger ener-gy barrier associated with the increased hop-ping distance to the next atom. Taking thelack of thermal diffusion across the rows inthe measurements of Briner et al. (18) tomean that it is at least three orders of magni-tude less probable than diffusion along therows, we then deduced a minimum barrierheight of 128 meV for diffusion across therows, compared to a barrier of 97 meV formotion along the rows. With this parameter,our analysis of an electronically driven pro-cess yields a branching ratio for diffusionacross the rows of 0.25. This value is com-parable to our experimental finding of 0.34but stands in sharp contrast to the equilibriumbehavior. The observed branching ratio re-flects the relative insensitivity to barrierheights of the electronically driven process.

These experiments show the possibility ofcombining direct imaging of molecules bySTM with access to the ultrafast time scale byfemtosecond laser excitation. The extensionof these measurements to multiple-pulse laserexcitation should retain the atomic-scale spa-tial resolution while providing further insightinto the temporal evolution of ultrafastdynamical processes.

References and Notes1. R. J. Hamers, D. G. Cahill, Appl. Phys. Lett. 57, 2031

(1990).2. S. Weiss, D. F. Ogletree, D. Botkin, M. Salmeron, D. S.

Chemla, Appl. Phys. Lett. 63, 2567 (1993).3. V. Gerstner, A. Knoll, W. Pfeiffer, A. Thon, G. Gerber,J. Appl. Phys. 88, 4851 (2000).

4. J. A. Prybyla, T. F. Heinz, J. A. Misewich, M. M. T. Loy,J. H. Glownia, Phys. Rev. Lett. 64, 1537 (1990).

5. J. A. Prybyla, H. W. K. Tom, G. D. Aumiller, Phys. Rev.Lett. 68, 503 (1992).

6. D. G. Busch, S. W. Gao, R. A. Pelak, M. F. Booth, W. Ho,Phys. Rev. Lett. 75, 673 (1995).

7. L. M. Struck, L. J. Richter, S. A. Buntin, R. R. Cavanagh,J. C. Stephenson, Phys. Rev. Lett. 77, 4576 (1996).

8. J. A. Misewich, S. Nakabayashi, P. Weigand, M. Wolf,T. F. Heinz, Surf. Sci. 363, 204 (1996).

9. R. J. Finlay, T. H. Her, C. Wu, E. Mazur, Chem. Phys.Lett. 274, 499 (1997).

10. G. Eichhorn, M. Richter, K. Al-Shamery, H. Zacharias,Chem. Phys. Lett. 289, 367 (1998).

11. M. Bonn et al., Science 285, 1042 (1999).12. D. N. Denzler, C. Frischkorn, C. Hess, M. Wolf, G. Ertl,Phys. Rev. Lett. 91, 226102 (2003).

13. K. Watanabe, N. Takagi, Y. Matsumoto, Phys. Rev.Lett. 92, 057401 (2004).

14. H. Petek, M. J. Weida, H. Nagano, S. Ogawa, Science288, 1402 (2000).

15. R. Gomer, Rep. Prog. Phys. 53, 917 (1990).16. T. T. Tsong, Prog. Surf. Sci. 67, 235 (2001).17. J. Ahner, J. T. Yates, J. Chem. Phys. 105, 6553 (1996).18. B. G. Briner, M. Doering, H.-P. Rust, A. M. Bradshaw,Science 278, 257 (1997).

19. L. Bartels, M. Wolf, G. Meyer, K. H. Rieder, Chem.Phys. Lett. 291, 573 (1998).

20. H. J. Lee, W. Ho, Science 286, 1719 (1999).21. A. J. Heinrich, C. P. Lutz, J. A. Gupta, D. M. Eigler,Science 298, 1381 (2002); published online 24 Oc-tober 2002; 10.1126/science.1076768.

22. T. Komeda, Y. Kim, M. Kawai, B. N. J. Persson, H. Ueba,Science 295, 2055 (2002).

23. J. I. Pascual, N. Lorente, Z. Song, H. Conrad, H. P. Rust,Nature 423, 525 (2003).

24. T. A. Germer, J. C. Stephenson, E. J. Heilweil, R. R.Cavanagh, Phys. Rev. Lett. 71, 3327 (1993).

25. T. A. Germer, J. C. Stephenson, E. J. Heilweil, R. R.Cavanagh, J. Chem. Phys. 101, 1704 (1994).

26. J. P. Culver, M. Li, Z. J. Sun, R. M. Hochstrasser, A. G.Yodh, Chem. Phys. 205, 159 (1996).

27. J. P. Culver, M. Li, R. M. Hochstrasser, A. G. Yodh, Surf.Sci. 368, 9 (1996).

28. M. Head-Gordon, J. C. Tully, Phys. Rev. B 46, 1853(1992).

29. C. Springer, M. Head-Gordon, Chem. Phys. 205, 73(1996).

30. J. T. Kindt, J. C. Tully, M. Head-Gordon, M. A. Gomez,J. Chem. Phys. 109, 3629 (1998).

31. D. A. Micha, Z. G. Yi, Chem. Phys. Lett. 298, 250(1998).

32. Our experiments were performed with a variable-temperature STM, which was operated at a basepressure of 8 10�11 mbar. The Cu(110) samplewas prepared by multiple cycles of sputtering (Ne�

at 1 keV) and annealing (300 s at 900 K).33. We gratefully acknowledge postdoctoral fellowships

from the Humboldt Foundation, the Novartis Foun-dation, and the Academia Leopoldina. This work wassupported by the U.S. Department of Energy, Officeof Basic Energy Sciences, through Catalysis Sciencegrant number DE-FG03-03ER15463/4; by the Amer-ican Chemical Society Petroleum Research Fund; andby equipment grants F49620-96-1-0406 andF49620-01-1-0286 from the U.S. Air Force Office ofScientific Research.

29 April 2004; accepted 15 June 2004Published online 24 June 2004;10.1126/science.1099770Include this information when citing this paper.

Nanoparticles: Strained and StiffBenjamin Gilbert,1 Feng Huang,1 Hengzhong Zhang,1

Glenn A. Waychunas,2 Jillian F. Banfield1,2*

Nanoparticles may contain unusual forms of structural disorder that can sub-stantially modify materials properties and thus cannot solely be considered assmall pieces of bulk material. We have developed a method to quantifyintermediate-range order in 3.4-nanometer-diameter zinc sulfide nanoparticlesand show that structural coherence is lost over distances beyond 2 nanometers.The zinc-sulfur Einstein vibration frequency in the nanoparticles is substantiallyhigher than that in the bulk zinc sulfide, implying structural stiffening. Thiscannot be explained by the observed 1% radial compression and must beprimarily due to inhomogeneous internal strain caused by competing relax-ations from an irregular surface. The methods developed here are generallyapplicable to the characterization of nanoscale solids, many of which mayexhibit complex disorder and strain.

The electronic properties of nanoparticles candiffer from those of their corresponding bulkform due to confinement effects caused onlyby their finite size, and because they arestructurally distinct. Quantum confinement isthe dominant size effect, and has been wellstudied (1–6). Structural deviations in nano-particles relative to bulk material are not wellunderstood because they are hard to resolveexperimentally (7). Consequently, theoreticalmodels of nanoparticles generally assumethey have bulklike interior structure (8).Tight-binding calculations that optimized

nanoparticle structure, and assumed full the-oretical passivation of surface anions, sug-gested that surfaces of nanoparticles relax ina manner comparable to that of bulk surfaces(9). However, classical and quantum molec-ular dynamics simulations have indicated thatdisorder may pervade throughout nanopar-ticles (10, 11). We previously showed thatnanoparticles can undergo substantial trans-formation in structure at low temperature,driven by surface interactions, indicating thatinternal strain depends upon the nature of thesurroundings as well as size (12). However, adetailed description of the strain within nano-particles has not been experimentally ob-tained. We combine pair distribution functionand extended x-ray absorption fine structure(EXAFS) analyses to quantify the structuraldistortion within mercaptoethanol-coatedzinc sulfide (ZnS) nanoparticles and the con-sequent changes in lattice dynamics.

1Department of Earth and Planetary Sciences, Univer-sity of California at Berkeley, Berkeley, CA 94720,USA. 2Earth Sciences Division, Lawrence Berkeley Na-tional Lab, One Cyclotron Road, Berkeley, CA 94720,USA.


R E P O R T S


Mercaptoethanol-coated ZnS nanopar-ticles were synthesized using the method ofVogel et al. (13). Fits to small-angle x-rayscattering (SAXS) data indicate that the av-erage diameter is 3.4 nm with a full width athalf maximum (FWHM) of 0.6 nm for thesize distribution (Fig. 1), in close agreementwith the size estimated from ultraviolet-visible absorption spectroscopy (fig. S1)(14). High-resolution transmission electronmicroscope (HRTEM) data confirm the sizedetermined by SAXS and show that the nano-particles are approximately spherical (fig. S2).

We acquired wide-angle x-ray scattering(WAXS) patterns from powders of bulk ZnS(sphalerite structure) and mercaptoethanol-coated ZnS nanoparticles (fig. S3). Althoughthe WAXS patterns from the nanoparticlesand from the bulk material have peaks inessentially the same positions (indicating thatthey share the same basic structure), thepeaks from the nanoparticles are wider andless intense. We obtained the real-space in-teratomic distance correlation functions,commonly known as pair distribution func-tions (PDFs), from the WAXS data usingstandard methods (14–17). The PDFs of bulkZnS and ZnS nanoparticles are given in Fig.2A. The PDF reveals interatomic correlationsat much greater distances than can beachieved by EXAFS (see below) (18–20). Incontrast to Bragg peak analysis of diffractiondata (21), the PDF additionally uses informa-tion from total diffuse x-ray scattering.

The nanoparticle PDF approaches zero at2.0 nm rather than at 3.4 nm, indicating thepresence of structural disorder. A detailedanalysis of the disorder in the nanoparticlescan be made by comparing the observed PDFto one obtained by truncating the PDF frombulk ZnS with a real-space curve associatedwith a 3.4-nm-diameter sphere (14, 22). Thisoperation creates the PDF of “ideal” ZnSsphalerite nanoparticles, also shown in Fig.2A, allowing finite particle-size effects on thediffraction data to be separated from structur-al effects.

The PDF for real ZnS nanoparticles isdistinct from that of ideal nanoparticles in thefollowing respects: (i) The first-shell PDFpeak intensity, representing Zn-S bondsthroughout the nanoparticles, is lower for realthan for ideal nanoparticles. (ii) PDF peakintensities at higher correlation distances di-minish more rapidly in the real nanoparticles.(iii) PDF peak widths are broader in the realnanoparticles. (iv) PDF peak positions areshifted. Observations (i) to (iii) indicate thepresence of two distinct forms of excess staticdisorder within the real nanoparticles relativeto bulk sphalerite (Fig. 3, A and B). The lowdependence on temperature of WAXS mea-surements (fig. S7) indicates that this excessdisorder is structural (i.e., static disorder)rather than vibrational (i.e., thermal disorder).The reduction and broadening of all peaks inthe PDF indicate that the random meansquared displacements about equilibrium po-sitions (uncorrelated with the positions ofneighboring atoms) are increased (Fig. 3A).However, this type of disorder broadens allpeaks in the PDF equivalently and cannot

explain the loss of intensity with increasinginteratomic distance. Thus, a second effect isthat strain within the nanoparticle causes cor-related shifts in the equilibrium positionsthemselves, relative to the perfect sphaleritelattice (Fig. 3B). This effect is cumulativewith increasing interatomic distance andcauses peak intensities in the real nanopar-ticle PDF to vanish at a correlation length ofabout 2.0 nm.

In addition, observation (iv) indicates acontraction of nanoparticle bond lengths (Fig.3C). We quantified the lattice contraction andthe two types of disorder by fitting real-spaceinteratomic correlations from the sphaleritestructure to the experimental PDFs (14).From a fit to the bulk sphalerite data (fig. S5),we obtained the thermal contribution to PDFpeak broadening by fitting the mean squaredrelative displacement (MSRD) while ac-counting empirically for correlated atomicmotion (23). From a fit to the nanoparticledata that accounted for the nanoparticle sizeand incorporated the thermal MSRD valueobtained from the bulk, we obtained two

Fig. 1. Small-angle x-ray scattering (SAXS) data(circles) from mercaptoethanol-coated ZnSnanoparticles in aqueous solution. The fit (solidline) is for polydisperse dilute spheres of aver-age diameter 3.4 nm and Schultz size distribu-tion FWHM of 0.6 nm.

Fig. 2. (A) PDFs of bulk ZnS (sphalerite, dashed gray curve), bulk ZnS following truncation by theshape factor for a 3.4-nm-diameter sphere (gray curve), and mercaptoethanol-coated ZnS nano-particles of average diameter 3.4 nm (black curve). (B) Theoretical fit (gray curve) to the PDF of ZnSnanoparticles (black curve), accounting for finite particle size and including the thermal meansquare relative displacement (MSRD) obtained from a fit to the bulk sphalerite data (inset,diamonds). In addition, the fit includes a static harmonic disorder contribution to the MSRD (inset,squares) and a short-range order (SRO) parameter (inset, circles). The MSRD contributions broadenthe PDF peaks; the SRO parameter is an intensity scaling factor.

R E P O R T S


additional static disorder parameters thathave a dependence on interatomic distance, r.The first parameter is the static MSRD con-tribution to PDF peak broadening due to ran-dom displacements about equilibrium posi-tions. The second is a short-range orderparameter, PSRO (r), where 0 � PSRO � 1,that causes the loss of PDF peak intensity dueto the correlated shifts in equilibrium posi-tions. The best-fit results are given in Fig. 2B.

The static MSRD contribution is shown inthe inset to Fig. 2B. Its form resembles that ofthe thermal MSRD (obtained from bulk ZnS)in that there is less disorder at short correla-tion lengths. For thermal disorder, this line-

shape is due to correlated atomic motion (23).By analogy, the static disorder is also corre-lated over short distances.

The best-fit curve PSRO (r) approacheszero with increasing correlation distance.This shows that, at larger interatomic distanc-es, fewer atoms maintain definite structuralrelationships relative to other atoms in thenanoparticle. This finding is consistent withpreservation of short-range order, strong re-duction in intermediate-range order, andcomplete loss of structural coherence at dis-tances greater than 2 nm (16).

From the PDF fit, we measured a mean 1%radial contraction of interatomic correlations inthe real nanoparticles. However, the peak shiftsare more complex, as indicated by a plot ofnanoparticle peak shift versus sphalerite peakposition (Fig. 3C). From studies on a modelnanoparticle subjected to structural modifica-tions, we can show that simple models for inter-nal strain can be distinguished by the shifts in theassociated PDF peak positions (fig. S8). Howev-er, we conclude that the strain within the realnanoparticles is more complex than that de-scribed by any of the simple models that weconsidered, including uniform, or surface-weighted, radial strain; linear strain; or the pres-ence of stacking faults.

The loss of structural coherence and theinability of simple strain models to describepeak shifts in PDF data point to the existenceof complex strain fields within the nanopar-ticles. We infer that the pervasive internaldistortion is due to strain that arises from thecompeting attempts of the structurally diverseterminating surfaces that encompass thenanoparticles to adopt lower energy configu-rations. Even with strong chemical passiva-tion, nanoparticle surfaces are generally non-stoichiometric [because capping ligandsinteract with surface cations or anions only,e.g. (20)] and incompletely capped [due tosteric limits, e.g. (24)]. The inherent limit of

surface passivation creates structural effectsthroughout the nanoparticle interior, despitethe highly crystalline appearance within thetransmission electron micrographs (fig. S2)or from x-ray diffraction (fig. S3). We antic-ipate that the observed forms of disorder aregeneral features of nanoscale solids and willbe especially pronounced where there is min-imal passivation.

Because internal strain and disorder cansubstantially affect materials properties, weinvestigated the lattice dynamics of the ZnSsamples. We obtained Zn K-edge EXAFSspectra between 3 and 500 K from the samebulk ZnS and ZnS nanoparticles used in thePDF analysis (Fig. 4A and fig. S9). TheEXAFS spectra from bulk ZnS contain peaksfrom long-range structure that are weak orabsent in the spectra from the nanoparticles,even at low temperature.

We obtained the first-shell (Zn-S) MSRD asa function of temperature by fitting the data inFig. 4A to ZnS scattering paths obtained throughsimulation using the FEFF code (14, 25). In Fig.4B, the nanoparticle MSRD curve is offset fromthat of the bulk, indicating the presence of excessstatic disorder in the nanoparticles. The room-temperature EXAFS and PDF analyses yieldconsistent measurements of this excess structuraldisorder. First-shell combined thermal and staticMSRDs in bulk and nanoparticles are 5.1 � 10–3

Å2 and 6.4 � 10–3 Å2, respectively, from fits tothe EXAFS data, and 5.0 � 10–3 Å2 and 6.6 �10–3 Å2 from fits to the PDF data.

From Fig. 4B, the amplitudes of vibrationsin bulk ZnS begin to increase rapidly withtemperature at �70 K, but this increase is notseen in the nanoparticles until �130 K. Thisindicates structural stiffening in the nanoparticles.

We fitted the temperature response of theMSRD to a theoretical expression for theanharmonic Einstein oscillator (14, 26). Inthe Einstein model, the Zn-S bonding pair isconsidered to vibrate in the center of mass

Fig. 4. (A) Fourier transform magnitude of k3-weighted Zn K-edge EXAFS spectra of bulk ZnS(sphalerite) and ZnS nanoparticles as a function of temperature. (B) Temperature dependence ofthe first-shell mean square relative displacement (MSRD) obtained from fits to the EXAFS data. Thelines are fits to the first-shell MSRD data using an anharmonic Einstein oscillator model to extractcharacteristic vibration frequencies for the two samples. A discussion of the errors in the MSRDdetermination is given in the Supporting Online Material (14 ).

Fig. 3. Structural modifications in ZnS nanopar-ticles can be decomposed into two distinctdisorder types, plus lattice contraction. (A)Random displacement disorder. Individual at-oms are randomly displaced from sites on asingle lattice. (B) Strain-driven distortion. Localstructure is maintained, but at larger inter-atomic distances, atoms lie farther from thepositions expected for an undistorted lattice.(C) Lattice contraction. The shift in PDF peakpositions observed in ZnS nanoparticles is plot-ted against the positions of the equivalentpeaks in the bulk sphalerite PDF.

R E P O R T S


frame under the influence of the interatomicpotential of the remaining stationary lattice.The characteristic vibration frequency ob-tained for bulk ZnS is 7.12 � 1.2 THz. Thisvalue is consistent with that reported previ-ously for bulk CdSe (27, 28). In comparison,the characteristic vibration frequency for theZnS nanoparticles is 11.6 � 0.4 THz. Al-though these values cannot be simply relatedto the full phonon dispersion relations forZnS, the trend clearly indicates lattice stiff-ening. This implies that the heat capacity ofthe ZnS lattice, excluding surface-bound spe-cies, is lower in nanoparticles than in equiv-alent bulk material.

There are several features of the nanopar-ticles that may be responsible for the latticestiffening. First, the PDF analysis indicates�1% bond length compression. Becausecrystalline materials show an increase in vi-bration frequencies with pressure (away fromphase transitions), the compression will causelattice stiffening. However, compression ofthis magnitude alone cannot explain the largeincrease in vibration frequency. The averagechange in bond length provides an incom-plete picture because disorder within thenanoparticles leads to a large distribution ofbond lengths, and all deviations from theequilibrium bond lengths may increase thevibration frequency. We thus find that struc-tural disorder is principally responsible forthe lattice stiffening.

In summary, we quantitatively determinednanoparticle crystallinity and disorder, usinga PDF-based method. We found that the bestdescription of nanoparticle structure includesbond length contraction, random disorder,and a type of disorder characterized by cor-related atomic displacements. The internaldisorder is observed despite the presence ofstrongly bound surface ligands. We concludethat even with strong chemical passivation,surface atoms exist in diverse unsatisfiedbonding environments at the surface thatdrive inhomogeneous internal strain. A strik-ing consequence of the internal disorder isthat the nanoparticles are much stiffer thanexpected from the measured bond length con-traction. Our approach is an important steptoward a realistic description of nanoparticlestructure that includes internal strain, whichis likely to be a general feature of nanoscalesolids. Because lattice contraction and disor-der will separately modify electronic proper-ties, inclusion of these effects is essential foraccurate nanoparticle calculations.

References and Notes1. R. Rossetti, R. Hull, J. M. Gibson, L. E. Brus, J. Chem.

Phys. 82, 552 (1985).2. C. B. Murray, D. J. Norris, M. G. Bawendi, J. Am. Chem

Soc. 115, 8706 (1993).3. A. P. Alivisatos, Science 271, 933 (1996).4. L. E. Brus, J. Chem. Phys. 80, 4403 (1984).5. P. E. Lippens, M. Lannoo. Phys. Rev. B 39, 10935

(1989).

6. A. Franceschetti, A. Zunger, Phys. Rev. Lett. 78, 915(1997).

7. B. Palosz et al., Z. Kristallogr. 217, 497 (2002).8. L.-W. Wang, A. Zunger, Phys. Rev. B 53, 9579 (1996).9. K. Leung, K. B. Whaley, J. Chem. Phys. 110, 11012

(1999).10. E. Rabani, J. Chem. Phys. 115, 1493 (2001).11. A. Puzder, A. J. Williamson, F. A. Reboredo, G. Galli,

Phys. Rev. Lett. 91, 157405 (2003).12. H. Zhang, B. Gilbert, F. Huang, J. F. Banfield. Nature

424, 1025 (2003).13. W. Vogel, P. H. Borse, N. Deshmukh, S. K. Kulkarni,

Langmuir 16, 2032 (2000).14. See supporting online material.15. B. J. Thijsse, J. Appl. Crystallogr. 17, 61 (1984).16. A. C. Wright et al., J. Non-Cryst. Solids 129, 213

(1991).17. S. J. L Billinge, M. F. Thorpe, Eds., Local Structure from

Diffraction (Plenum, New York, 1998).18. J. Rockenburger, et al. J. Chem. Phys. 108, 7807

(1998).19. M. A. Marcus, W. Flood, M. Steigerwald, L. Brus, M.

Bawendi. J. Phys. Chem. 95, 1572 (1991).20. A. C. Carter, et al., Phys. Rev. B 55, 13822 (1997).21. M. G. Bawendi, A. R. Kortan, M. L. Seigerwald, L. E.

Brus, J. Chem. Phys. 91, 7282 (1989).22. G. Mason, Nature 217, 733 (1968).23. I.-K. Jeong, Th. Proffen, F. Mohiuddin-Jacobs, S. J. L.

Billinge, J. Phys. Chem. A 103, 921 (1999).24. J.-J. Shiang, A. V. Kadavanich, R. K. Grubbs, A. P.

Alivisatos, A. P. 1995. J. Phys. Chem. 99, 17417(1995).

25. A. L. Ankudinov, B. Ravel, J. J. Rehr, S. D. Conradson,Phys. Rev. B 58, 7565 (1998).

26. A. I. Frenkel, J. J. Rehr, Phys. Rev B 48, 585 (1993).27. G. Dalba, et al., Phys. Rev. B 58, 4793 (1998).28. For two atoms vibrating in their center of mass frame

in an effective pair potential characterized by springconstant k, the bond frequency, � � �k/ , where is the reduced mass. For CdSe, 1/ � 1/MCd �

1/MSe (MCd is the atomic mass of Cd). For CdSe, CdSe�4.97 THz (27). For ZnS, we obtained ZnS � 7.12THz.The interatomic force constants for CdSe and ZnS areexpected to be similar. If we assume that kCdSe �

kZnS, then�CdSe

�ZnS predicted

��ZnS

CdSe� 0.68. In fact,

�CdSe

�ZnS observed

��ZnS

CdSe�0.676. Hence, the results pre-

sented here for bulk ZnS and in (27 ) for bulk CdSe arein good agreement.

29. We thank M. Toney for discussion on PDF analysis, C.Kim for practical advice on EXAFS acquisition, and M.Zach for assistance with sample holder fabrication.SAXS data were acquired on beamline 1.4 at theStanford Synchrotron Radiation Lab (SSRL), and wethank J. Pople. Zn K-edge EXAFS were acquired onbeamline 4-3 at SSRL, and we thank M. Lattimer.WAXS data were acquired on beamline 11-ID-C atthe Advanced Photon Source (APS), Argonne Nation-al Laboratory, and we thank Y. Ren and M. Beno. TheU.S. Department of Energy, Office of Basic EnergySciences, supports use of the APS (Contract No.W-31-109-Eng-38) and the SSRL (Contract No. DE-AC03-76SF00515). HRTEM was performed at theNational Center for Electron Microscopy, Berkeley,CA, and we thank C. Song. Financial support for thiswork came from the U.S. Department of Energy, theNational Science Foundation, and Lawrence BerkeleyNational Laboratory.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/1098454/DC1Materials and MethodsFigs. S1 to S9References

29 March 2004; accepted 22 June 2004Published online 1 July 2004;10.1126/science.1098454Include this information when citing this paper.

Grain Boundary–MediatedPlasticity in Nanocrystalline Nickel

Zhiwei Shan,1 E. A. Stach,2 J. M. K. Wiezorek,3 J. A. Knapp,4

D. M. Follstaedt,4 S. X. Mao1*

The plastic behavior of crystalline materials is mainly controlled by the nucleationandmotion of lattice dislocations.We report in situ dynamic transmission electronmicroscope observations of nanocrystalline nickel films with an average grain sizeof about 10 nanometers, which show that grain boundary–mediated processeshave become a prominent deformation mode. Additionally, trapped lattice dislo-cations are observed in individual grains following deformation. This change in thedeformationmode arises from the grain size–dependent competition between thedeformation controlled by nucleation and motion of dislocations and the defor-mation controlled by diffusion-assisted grain boundary processes.

The plastic deformation of coarse-grainedmetals at relatively low temperature is usual-ly mediated by the nucleation and motion of

dislocations. However, for nanocrystallinemetals, dislocation sources and pile-ups arenot expected to exist within the individualgrains because of the limited grain size andthe much higher fraction of grain boundary(GB) atoms (1–8). It has been proposed (3–6)that GB-mediated plasticity (e.g., GB slidingand/or grain rotation) substitutes for conven-tional dislocation nucleation and motion asthe dominant deformation mechanism whengrain sizes are reduced below a certain value.This conjecture has been supported by mo-lecular dynamics simulations (5, 7, 8) andindirect experimental evidence (2, 9). Dy-

1Department of Mechanical Engineering, University ofPittsburgh, 648 Benedum Hall, Pittsburgh, PA 15261,USA. 2National Center for Electron Microscopy(NCEM), Lawrence Berkeley National Laboratory(LBNL), Berkeley, CA 94720, USA. 3Department ofMaterials Science and Engineering, University of Pitts-burgh, Pittsburgh, PA 15261, USA. 4Physical andChemical Science Center, Sandia National Laborato-ries (SNL), Albuquerque, NM 87185, USA.


R E P O R T S


namic in situ deformation studies (10–13)have been performed in the transmissionelectron microscope (TEM) to observe thedeformation mechanisms that are active innanocrystalline metals with grain sizes in theregime where a transition in mechanism ispredicted. However, direct experimental evi-dence of a transition in deformation mecha-nisms, i.e., from a dislocation-mediated de-formation to a GB-mediated deformation, hasnot been reported (10–14).

The Ni samples used in this study weresynthesized by directing a high-energy pulsedKrF excimer laser onto a high-purity Ni tar-get under vacuum with a base pressure ofabout 2 � 107 torr. The resulting Ni plasmawas deposited onto a [001] NaCl substratewith a nominal thickness of 150 nm. TEMobservations (Fig. 1A) indicate that the as-deposited Ni consists of roughly equiaxedgrains with random orientations (inset in Fig.1A). Statistical measurements of dark-fieldTEM (DFTEM) images reveal a narrow, log-normal grain size distribution, ranging fromseveral nanometers to 23 nm with an averagevalue (�SD) of 9.7 � 3.9 nm (Fig. 1B).High-resolution TEM (HRTEM) shows thatmost grains are separated by large-angleGBs. Additional GB phases, porosity, or in-tergranular microcracks were not detected inthese samples after deposition.

The nanocrystalline nickel films weredeformed by in situ TEM tensile straining(15). Figure 2A is a typical DFTEM imageof a Ni sample in the undeformed state.Figure 2B is the corresponding selectedarea diffraction pattern (SADP) before de-formation. Upon straining, bright-fieldTEM observations showed rapid changes incontrast that occur continuously in manydifferent grains. This has previously beenidentified as evidence of dislocation activ-ity (11, 13). However, our in situ DFTEMobservations revealed that the deformationbehavior is substantially more complex.We found many strongly diffracting fea-tures that are much larger than the initialaverage grain size and are formed immedi-ately in the deformed zone upon straining(e.g., the feature marked by a white arrowis about 60 nm in diameter in Fig. 2C). TheSADP from the deformed area (Fig. 2D,taken from the same region and with thesame selected area aperture size as the areain Fig. 2B) exhibited fewer, asymmetricallydistributed diffraction spots in each of thediffraction rings in comparison with theSADP obtained before deformation (Fig.2B). The reduced background intensity andimproved contrast in the SADP of the de-formed area (Fig. 2D) indicate that thesample thinned locally during deformation.Higher magnification DFTEM micrographsrevealed that the large bright features ob-served after deformation consist of a num-

ber of smaller grains (e.g., Fig. 2E) ratherthan a single large grain resulting fromstress-assisted grain growth. A comparisonof the undeformed state (Fig. 2, A and B)and the postdeformation state (Fig. 2, C, D,and E) indicates that groups of neighboringgrains underwent an orientation changeduring straining and formed numerousgrain agglomerates. Because of the natureof the image formation mechanism ofDFTEM images (15), we can conclude thatthe smaller grains in these agglomeratesexhibit essentially edge-on orientations oftheir {111} and/or {200} lattice planes.The schematic in Fig. 2F depicts a possiblecrystallographic substructure associatedwith an agglomerated group of grains thatwould be consistent with the type of con-trast observed in the DFTEM micrographsafter straining (Fig. 2E).

To elucidate further the mechanism re-sponsible for the formation of the agglomer-ated grain regions, real-time observationswere performed. Contrast changes in areassubject to high strain during in situ TEMstraining experiments were recorded in thedark-field (DF) mode. The DF micrographs

shown in Fig. 3 are still frames extractedfrom a typical dynamic sequence of imagestaken during the application of a single dis-placement pulse. The times listed on eachstill frame are based on the video-acquisitionrate of 30 frames per second. At the begin-ning of this sequence, there were no grains ina strongly diffracting condition in the areaindicated by the white arrow (Fig. 3A). After1/30 of a second, a bright spot emerged froma grain about 6 nm in diameter and remainedwell defined in size as a single, approximate-ly equiaxed grain until t � 0.1 s (Fig. 3B).Over the next couple of frames, a number ofadditional neighboring grains rotated intostrongly diffracting conditions for either the{200} or {111} planes. Other nearby grains,which were in a strong diffraction condition,did not rotate out of contrast, confirming thatthere had been no global rotation of the spec-imen area and that the rotations observed at

Fig. 1. TEM observations of the typical micro-structure in the as-deposited nanocrystallinenickel films. The bright-field TEM micrograph (A)and the SADP [inset in (A)] show roughly equi-axed grains with random orientations. The statis-tical distributions for grain size (B) were obtainedfrom multiple TEM images of the same sample.

Fig. 2. TEM micrographs showing the evolutionof the Ni microstructure during in situ straining.(A) DFTEM image of as-deposited nanocrystal-line Ni film before deformation. Part of the{111} and {200} diffraction rings were selectedas the image-forming diffraction vectors, asindicated schematically by the white circle in(B) (all of the DF images here are taken withthis condition). (B) Corresponding SADP of un-deformed Ni film. (C) DFTEM micrograph ofas-deposited Ni after deformation. Features instrong diffraction contrast much larger thanthe average grain size are discernible (e.g.,white arrow). (D) The corresponding SADP ofthe deformed Ni film. (E) Higher magnificationDFTEM image of the feature marked by thewhite arrow in (C). (F) Schematic sketch of apossible multigrain substructure of the largefeatures in bright contrast observed after de-formation [e.g., in (C)]. The lines represent thetrace of those edge-on planes.

R E P O R T S


the arrowed location were internal changes ofthe sample structure. Additionally, the small“notch” discernible at the lower left cornerof the agglomerate (Fig. 3, D and E) indi-cates that the growth in size of this agglom-erated group of grains was not isotropic andinvolved sudden rotation of individualgrains. At t � 0.5 s, the group of grains hadgrown into an elongated equiaxed shapewith an approximate size of 60 nm alongthe short axis and 80 nm along the longaxis. After this very rapid morphologicalchange, the rate of growth of the grainagglomerate decreased significantly.

These TEM observations confirm the pre-diction that GB-mediated deformation canbecome prominent when the average grainsize of a material decreases below some crit-ical value. Several theoretical studies haveconsidered the operation of grain rotation andGB sliding as possible deformation modes(16–18). These studies assume that the driv-ing force for grain rotation is the net torqueon a grain, which results from the misorien-tation dependence of the energy of the GBsthat delineate a given grain from its neigh-bors. Grain rotation is viewed as a slidingproblem on the periphery of the grain; chang-es in the grain shape during rotation areassumed to be accommodated by diffusion,through either the GBs or the grain interiors.If GB diffusion is dominant (a reasonableassumption in nanocrystalline metals withgrain size below a certain value at roomtemperature), it would yield a d–4 depen-

dence on the rotation rate, where d is thegrain size. This is a marked dependence ongrain size. For example, when d is 60 and 6nm, respectively, and all other factors areidentical, the grain rotation rate for the latterwill be 104 higher than for the former. Thisqualitatively explains our observation of ex-tremely rapid formation of the grain agglom-erates by rotation during straining. The rate ofthe grain rotation processes will generallydecrease with time, as a state of new equilib-rium is approached locally.

Previous TEM observations have indicat-ed that dislocations may still play a role indeformation even if the average grain sizefalls below the grain size at which GB pro-cesses may become active (13). Hence, itmay be expected that evidence of lattice dis-location glide activity during straining wouldexist in the microstructure of deformed nano-structured metals. However, detection of suchevidence by postmortem TEM investigations,for instance in the form of dislocation debrisor trapped lattice dislocations, has provenunsuccessful (19). Yamakov et al. (20) sug-gested that trapped dislocations may onlyexist inside a nanograin under very high ex-ternal or internal stress and that the removalof the stress can lead to reversible reabsorp-tion of the partial dislocations at the pre-sumed GB source. However, the traditionalsample preparation methods, such as me-chanical thinning followed by ion thinning ortwin jet polishing will inevitably result insome relaxation of a previously deformedmicrostructure. This makes it difficult to ver-ify the assumption, and therefore limits fur-ther understanding of this important funda-mental process. During this study, HRTEMimages of suitably oriented grains that werestill under stress were obtained in the thinarea that was produced by the deformation.

Trapped lattice dislocations were detected insome of the grains. Figure 4 shows an exam-ple of an HRTEM micrograph of one of thesetrapped dislocations (white T) trapped in thevicinity of a GB (delineated by the blackdashed line). The inset at the upper rightcorner in Fig. 4 is an inverse Fourier-filteredimage of the region framed by the white box,which displays the dislocation more clearly.Local Burgers circuits were employed to de-termine the Burgers vectors of the disloca-tions in HRTEM. As for Fig. 4, the Burgersvector of the trapped dislocation was deter-

mined to be consistent with b �a

2110�,

where a is the lattice constant of Ni—i.e., aunit dislocation of the face-centered cubiclattice on a (1�11) plane (15). The frequentobservation of trapped lattice dislocationsin the postdeformation state indicates thatdislocation-mediated deformation is stillactive even when the average grain size isabout 10 nm. Two other recent experimentshave also detected the presence of disloca-tions in nanocrystalline Ni while understress (21, 22), although for larger grains(20 and 26 nm, respectively).

The physics governing the observed defor-mation crossover can be understood by consid-ering the effect of grain size on the differentoperative processes. A number of computationalsimulations have predicted that GBs can act asdislocation sources in nanocrystalline materials(5, 23, 24). Based on these simulations, Chen etal. (25) proposed a dislocation-based model sug-gesting that the nucleation stress for both perfectand partial dislocations is inversely proportionalto the grain size. Furthermore, this model pre-dicts the existence of a critical grain size dc,below which the deformation mechanism willchange from one controlled by normal unit dis-location motion to one controlled by partial dis-location activity. If we take the dislocation coreparameter � � 1, the shear modulus as �Ni � 95GPa (26), and the stacking fault energy of nickelas �0.128 to 0.24 J/m2 (27, 28), the critical size(dc) for Ni then ranges from �11 to 22 nm.However, according to this theory, the lowerbound of the nucleation stress for partial dislo-cations (for d � 23 nm) in our sample is as highas �2.1 to 2.8 GPa. This indicates that very highlocal stress is necessary for the nucleation ofpartial dislocations in nanocrystalline Ni with anaverage grain size of 10 nm. In contrast, the rateof GB-mediated deformation, as mentionedabove, increases rapidly with a scaling of d–4.Thus, the deformation mechanism crossover isan inevitable result of the competition betweenthe deformation controlled by nucleation andmotion of dislocations (unit and partial) and thedeformation controlled by GB-related deforma-tion accommodated mainly by GB diffusionwith decreasing grain size.

In a critical part of our investigation, we

Fig. 3. DFTEM observation of the rapid genesis ofan agglomerate (e.g., white arrow) depicted byindividual still frames extracted from a dynamicvideo sequence. (A) At t � 0 s, no grains in thestrong diffraction condition are near the whitearrow. (B) At t � 0.1 s, a grain in the strongdiffraction condition with a size of about 6 nm isvisible. (C) At t � 0.2 s, a group of grains in brightcontrast with a size of about 28 nm is visible. (D)At t � 0.3 s, the group of grains has a nearlyelliptical shape, with dimensions of 60 by 35 nm.(E) At t � 0.4 s and (F) t � 0.5 s, the size of thegroup of grains increases to maximum dimen-sions of about 80 by 60 nm.

Fig. 4. A typical HRTEM image of a thin areaformed by deformation. A dislocation (white T ) istrapped inside a grain close to the GB (delineatedby black dashed line). The inverse Fourier-filteredimage (inset) from inside the white box showsthe dislocation (black arrowhead) with more clarity.

R E P O R T S


observed the rapid sequence of initial grainrealignments by closely examining succes-sive video frames in DFTEM mode. Thefrequent observation of the GB-mediated de-formation reported here (Figs. 2 and 3) wouldnot have been possible in bright-field TEMconditions because of the inherent difficultywith differentiating the contrast changescaused by GB-related deformations fromthose caused by the motion of lattice disloca-tions in small grains (e.g., less than 20 nm).

References and Notes1. J. R. Weertman et al.,Mater. Res. Soc. Bull. 24, 44 (1999).2. C. C. Koch, D. G. Morris, K. Lu, A. Inoue, Mater. Res.

Soc. Bull. 24, 54 (1999).3. H. Van Swygenhoven, Science 296, 66 (2002).4. J. R. Weertman, in Nanostructured Materials: Process-

ing, Properties and Applications, C. C. Koch, Ed. (Wil-liam Andrews Publishing, Norwich, New York, 2002),pp. 397–421.

5. J. Schiotz, F. D. DiTolla, K. W. Jacobsen, Nature 391,561 (1998).

6. S. Yip, Nature 391, 532 (1998).7. J. Schiotz, K. W. Jacobsen, Science 301, 1357 (2003).8. V. Yamakov, D. Wolf, S. R. Phillpot, A. K. Mukherjee,H. Gleiter, Nature Mater. 3, 43 (2004).

9. C. A. Schuh, T. G. Nieh, T. Yamasaki, Scripta Mater.46, 735 (2002).

10. M. Ke, S. A. Hackney, W. W. Milligan, E. C. Aifantis,Nanostructured Mater. 5, 689 (1995).

11. C. J. Youngdahl, J. R. Weertman, R. C. Hugo, H. H.Kung, Scripta Mater. 44, 1475 (2001).

12. K. S. Kumar, S. Suresh, M. F. Chisholm, J. A. Horton, P.Wang, Acta Mater. 51, 387 (2003).

13. R. C. Hugo et al., Acta Mater. 51, 1937 (2003).14. H. V. Swygenhoven, J. R. Weertman, Scripta Mater.

49, 625 (2003).15. Materials and methods are available as supporting

material on Science Online.16. R. Raj, M. F. Ashby, Metal. Trans. 2, 1113 (1971).17. K. E. Harris, V. V. Singh, A. H. King, Acta Mater. 46,

2623 (1998).18. D. Moldovan, D. Wolf, S. R. Phillpot, Acta Mater. 49,

3521 (2001).19. M. Legros, B. R. Elliott, M. N. Rittner, J. R. Weertman,

K. J. Hemker, Philos. Mag. A 80, 1017 (2000).20. V. Yamakov, D. Wolf, M. Salazar, S. R. Phillpot, H.

Gleiter, Acta Mater. 49, 2713 (2001).21. R. Mitra, W.-A. Chiou, J. R. Weertman, J. Mater. Res.

19, 1029 (2004).22. Z. Budrovic, H. Van Swygenhoven, P. M. Derlet, S. V.

Petegem, B. Schmitt, Science 304, 273 (2004).23. V. Yamakov, D. Wolf, S. R. Phillpot, A. K. Mukherjee,

H. Gleiter, Nature Mater. 1, 45 (2002).24. H. V. Swygenhoven, P. M. Derlet, A. Hasnaoui, Phys.

Rev. B 66, 024101 (2002).25. M. W. Chen et al., Science 300, 1275 (2003).26. J. P. Hirth, J. Lothe, Theory of Dislocations (Wiley,

New York, ed. 1, 1982), p. 837.27. L. E. Murr, Interfacial Phenomena in Metals and Alloys

(Addison Wesley, Reading, MA, 1975).28. P. S. Dobson, P. J. Goodhew, R. E. Smallman, Philos.

Mag. 16, 9 (1967).29. Supported by NSF grant CMS-0140317 to the Uni-

versity of Pittsburgh. The work at NCEM/LBNL and atSNL was supported by the Director, Office of Science,Office of Basic Energy Sciences, Division of MaterialsSciences and Engineering, of the U.S. Department ofEnergy. Sandia is a multiprogram laboratory operatedby Sandia Corporation, a Lockheed Martin Company,for the United States Department of Energy’s Na-tional Nuclear Security Agency under contract DE-AC04-94AL85000. We thank L. Lu and C. Y. Song forfruitful discussion.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/654/DC1Materials and MethodsFig. S1


Pinpointing the Source of a LunarMeteorite: Implications for theEvolution of the Moon

Edwin Gnos,1* Beda A. Hofmann,2 Ali Al-Kathiri,1

Silvio Lorenzetti,3 Otto Eugster,3 Martin J. Whitehouse,4

Igor M. Villa,1 A. J. Timothy Jull,5 Jost Eikenberg,6

Bernhard Spettel,7 Urs Krahenbuhl,8 Ian A. Franchi,9

Richard C. Greenwood9

The lunar meteorite Sayh al Uhaymir 169 consists of an impact melt brecciaextremely enriched with potassium, rare earth elements, and phosphorus [tho-rium, 32.7 parts per million (ppm); uranium, 8.6 ppm; potassium oxide, 0.54weight percent], and adherent regolith. The isotope systematics of the mete-orite record four lunar impact events at 3909 � 13 million years ago (Ma),�2800 Ma, �200 Ma, and 0.34 Ma, and collision with Earth sometime after9.7 � 1.3 thousand years ago. With these data, we can link the impact-meltbreccia to Imbrium and pinpoint the source region of the meteorite to theLalande impact crater.

The elevated Th content of the lunar Procel-larum terrane was recognized during theApollo gamma-ray remote mapping program(1). The terrane, which includes Mare Im-brium and Mare Procellarum, is characterizedby K-REE-P (or KREEP) [potassium (K),rare earth elements (REE), and phosphorus(P)] rock, which is enriched in incompatibleelements (2–5). Such material was returnedby all Apollo missions (6). KREEP-rich ma-terial is confined to areas surrounding theImbrium basin and the Montes Carpatus–Lalande belt (7), where the impact depositsprovide a datable marker of lunar stratigraphy(6, 8). The age of the Imbrium basin has beeninferred from isotope-system shock-resettingdata and is debated to be either 3770 or 3850million years old (9, 10).

Sayh al Uhaymir (SaU) 169 is a 206.45-grock found in the Sultanate of Oman (11).The rock consists of two lithologies (Fig. 1).About 87% by volume (estimates based on

tomographic sections) (fig. S1) consists of aholocrystalline, fine-grained polymict im-pact-melt breccia (stage I in Fig. 1) contain-ing 25 to 40 vol % of shocked rock andmineral clasts. The rock clasts are coarse-grained norites, gabbronorites, and maficgranulites. Crystal clasts comprise plagio-clase (An57-94), orthopyroxene (En44-78

Wo1-7), olivine (Fo58-67), and minor ilmenite,clinopyroxene, spinel, tridymite, or kamacite(table S1). The mineral chemistry of clastssuggests the presence of norites to olivinenorites and subordinate, more evolved mag-matites (granodiorites to granites) and a fewgranulites at the impact area. No highlandanorthosites (An�97) were identified.

The fine-crystalline impact melt (generally200-�m grain diameter) consists of short-prismatic, low-Ca pyroxene (En61-64Wo3-4),mildly shocked plagioclase (An75-81), and po-tassium feldspar, ilmenite, whitlockite, olivine(Fo58-59), zircon, troilite, kamacite, and tridym-ite (see also table S1). The impact melt il-menites contain unusually high Nb2O5 (�0.5wt %), which distinguishes them from knownlunar ilmenites (6). The impact melt is partiallyrimmed by shock-lithified regolith (13 vol%).We distinguish two stages of regolith formation(II and III in Fig. 1) from compositional differ-ences. Regolith II and III comprise clasts ofcrystalline and glassy volcanic rocks, magmaticrocks, breccias, mafic granulites, and crystalfragments. Only stage III regolith is borderedby flow-banded glass and contains yellow andorange glass fragments (including glass beads),olivine basalts, pyroxferroite-bearing basalts,and an anorthosite clast. The basaltic clastsencompass the full range of compositionsfrom Ti-rich to Ti-poor basalts, includingaluminous members and picrobasalts (tableS2). Glassy shock veins (stage IV in Fig. 1)

1Institut fur Geologie, Universitat Bern, Baltzerstrasse1, CH-3012 Bern, Switzerland. 2Naturhistorisches Mu-seum der Burgergemeinde Bern, Bernastrasse 15, CH-3005 Bern, Switzerland. 3Physikalisches Institut,Abteilung fur Weltraumforschung und Planetologie,Universitat Bern, Sidlerstrasse 5, CH-3012 Bern, Swit-zerland. 4Laboratory for Isotope Geology, SwedishMuseum of Natural History, Box 50007, SE-104 05,Stockholm, Sweden. 5National Science Foundation–Arizona Accelerator Mass Spectrometry Laboratory,University of Arizona, 1118 East Fourth Street, Tuc-son, AZ 85721, USA. 6Paul Scherrer Institut, 5232Villigen, Switzerland. 7Max-Planck-Institut fur Che-mie, Abteilung Kosmochemie, 55020 Mainz, Germa-ny. 8Departement fur Chemie und Biochemie, Univer-sitat Bern, Freiestrasse 3, CH-3012 Bern, Switzerland.9Planetary and Space Sciences Research Insti-tute, Open University, Milton Keynes MK7 6AA,UK.

*To whom correspondence should be addressed:[email protected]

R E P O R T S


crosscut impact melt and regolith and recordthe latest impact event.

The chemical composition of the meteor-

ite has been characterized by using comple-mentary methods. Nondestructive �-ray spec-troscopy of the uncut rock yielded a typicallylunar K/U ratio of 554. The mean of threeoxygen isotope measurements of the impactmelt (�17O � 3.37‰, �18O � 6.48‰, and17O � 0.001 � 0.032‰) plots on the Earth-Moon fractionation line (12). Chemical anal-yses were obtained from the impact melt, theregolith, and a large breccia clast (Table 1and Fig. 1). Th, U, and K concentrations aresimilar for powder aliquots of the impact meltand for the bulk rock. A lunar origin for themeteorite is indicated by lunar element ratios(6) such as Fe/Mn (moon, 60 to 80; impactmelt, 79; regolith, 78; regolith clast, 74) andK/U ratios (moon, 320 to 3800; impact melt,535; regolith, 1682; clast, 1253).

The SaU 169 impact-melt breccia ischemically similar to Apollo KREEP impact-melt breccias (4), with slightly lower Al andSi, and higher Na, Ti, and P. However, with32.7 ppm Th, 8.6 ppm U, and 1332 ppm

REEtot, the impact melt is more enriched inKREEP elements than is any other knownlunar rock (6). Mean concentrations in Apol-lo 14 to 17 Th-rich impact-melt brecciasrange from 8.2 to 16.7 ppm (13). Relative toApollo KREEP-rich impact melts, it is en-riched in incompatible trace elements by afactor of 2 to 4 (Fig. 2). The alkali elementsK, Rb, and Cs occur at levels typical forApollo impact melts but are depleted relativeto other incompatible trace elements: K/Th,Rb/Th, and Cs/Th are only 0.35 � 0.02 of theApollo impact-melt breccia values (Fig. 2),which indicates a decoupling of the alkalisduring incompatible trace-element fraction-ation. A low Rb/Sr of 0.04 shows that it is notrelated to highly evolved KREEP differenti-ates (Rb/Sr, 0.10 to 0.56) (13). The sidero-phile elements Ni, Co, Ir, and Au are presentat levels similar to Apollo impact melts andrequire �0.5% of a meteoritic component.

The average regolith (Fig. 1 and Table 1)has a composition similar to Apollo 12, 14,

Fig. 1. Thin-section photograph showing the different lithologies and their age relationship: I,impact-melt breccia. II and III, two types of regolith (III intrudes II). IV designates shock veins asyoungest formation. Ol, olivine; Plg, plagioclase; Opx, orthopyroxene. Image width is 35 mm.

Fig. 2. Element abundances of impact-melt breccia, regolith II and III com-bined, and a polymict breccia clast inregolith II normalized to averageApollo 14 high-Th impact-melt brec-cias (13). Data for Apollo 15 to 17impact melts are shown for compari-son. Compared with Apollo data, P, Zr,REE, Th, and U contents are signifi-cantly higher, but K, Cs, and Rb, aswell as Al and Si (typically fractionat-ed in potassium feldspar), are lower.

Table 1. Whole-rock chemistry.

Oxides orelements

Impact-meltbreccia827 mg*

Averageregolith185 mg†

KREEPclast

117 mg†

wt %SiO2 45.15 46.9‡ –Al2O3 15.88 17.54 16.34FeOtot 10.67 11.09 8.80MnO 0.14 0.14 0.12MgO 11.09 7.94 6.92CaO 10.16 11.72 10.60Na2O 0.98 0.78 1.18K2O 0.54 0.46 0.88TiO2 2.21 2.49 1.47ZrO2 0.38 0.08 0.19BaO 0.17 0.07 0.15Cr2O3 0.14 0.19 0.12P2O5 1.14 0.42 0.76S 0.33Total 98.98 99.82

ppmSc 25 28 18V 36 61 36Cr 992 1310 811Co 31 19 12Ni 204 82 58Cu 9 �20 �20Zn 31 �60 �60Rb 13.7 10.0 20.0Sr 359 214 230Y 532 162.5 338Zr 2835 596 1397Nb 124 18 112Cs 0.8 0.4 0.9Ba 1520 593 1351La 170 52 113Ce 427 139 297Pr 57.45 17.1 35.6Nd 256.5 76.9 162Sm 70.15 21.9 44.9Eu 4.20 2.43 2.45Gd 86.4 25.3 50.4Tb 15.1 5.08 10.5Dy 94.15 30.7 63.9Ho 21.3 6.36 13Er 58.05 18.6 39.3Tm 9.13 2.72 5.96Yb 54.65 16.9 36Lu 7.64 2.53 5.24Hf 64.3 14.8 34.7Ta 7.1 2.14 4.16W 3.45 1.3 2.5Ir ppb 4.2Au ppb 6Pb 13.8 �10 �10Th 32.7 8.44 21.70U 8.6 2.27 5.83

Sum REE (ppm) 1332 418 879Fe/Mn (wt) 76.5 79.5 73.6Mg/(FeMg) (mol) 0.65 0.56 0.58K/U (wt) 521 1682 1253Th/U (wt) 3.80 3.72 3.72

*Combined inductively coupled plasma mass spectrometryand optical emission spectrometry, and instrumental neu-tron activation analysis. †Inductively coupled plasmamass spectrometry and optical emission spectrometry.‡Microprobe value on regolith shock glass.

R E P O R T S


and 15 regoliths. In contrast to the impactmelt, the regolith is not depleted in Na and K(Fig. 2), as shown by the higher K/U of 1682.The regolith clast is an evolved KREEP rock(Fig. 2) with 21.7 ppm Th, but it is unrelatedto the impact melt, on the basis of its K/Uratio of 1253. The exceptionally high Th, U,and K contents restrict the provenance of SaU169 to the Imbrium area (13).

We determined the ages of the four im-pact events using isotopic data. 207Pb-206Pbisotope ratios (table S3) were obtained inthin section by ion-microprobe analysis on12 poikilitic impact-melt zircons and yield-ed a crystallization age of 3909 � 13 Ma[2�; mean square weighted deviation(MSWD), 0.33] (Fig. 3). 39Ar-40Ar data(table S4) were obtained on a feldspar con-centrate from the impact melt. The irregularage spectrum in combination with the Ca/Kconcentrations indicate a resetting of theCa-rich feldspars at �2800 Ma and ayounger disturbance at �500 Ma affectingthe potassium feldspars (fig. S2).

The duration of cosmic-ray exposure(CRE) of the impact-melt breccia and theyounger regolith (stage III in Fig. 1) wasobtained from light noble gas isotopic char-acteristics (table S5). All three samples ana-lyzed show He loss; hence, the 3He CRE ageis not significant. The production rates for21Ne and 38Ar depend on the chemical com-position and on shielding depth during expo-sure to cosmic rays. The (22Ne/21Ne)c valueof 1.197 for the impact melt and 1.200 for theregolith indicate irradiation at a shieldingdepth of a few tens of centimeters. Using themethod for calculating production rates pro-posed by (14) at the lunar surface and (15)during transfer in space, and assuming a typ-ical shielding of 40 g/cm2, we obtained lunarsurface CRE ages from 21Ne and 38Ar of200 � 40 Ma and 182 � 36 Ma for theimpact melt, and 150 � 30 Ma and 192 � 38Ma for the regolith. The 38Ar CRE age ex-tracted from the 39Ar-40Ar data is also 192 �20 Ma (table S4 and fig. S2).

By adopting a saturation activity for 10Beof 25 dpm/kg as given by (16), we found the

upper time limit for a small rock exposureduring Moon-Earth transfer, estimated from10Be (table S6), to be 0.34 Ma. Thus, theexposure in free space is negligibly shortcompared with the total CRE. We concludethat the CRE ages are lunar regolith residencetimes. Like most lunar meteorites (17), SaU169 thus seems to have resided in the uppermeter of the regolith before launch into space,as supported by our CRE ages and by thesolar wind component found in the stage IIIregolith (table S5). The terrestrial age ob-tained by 14C and 10Be methods is �9700 �1300 years (table S6).

The combined age data indicate a com-plex lunar history. Crystallization of theimpact melt occurred at 3909 � 13 Ma,followed by exhumation by a second im-pact at �2800 Ma, which raised the sampleto a regolith position at unconstraineddepth. A third impact at �200 Ma movedthe material closer to the lunar surface,where it mixed with solar-wind–containingregolith. It was launched into space by afourth impact at �0.34 Ma.

The 3909 � 13 Ma 207Pb-206Pb zirconage on the impact melt is an accurate estimatefor the Imbrium impact event. The smalldifference from the 3770 � 20 or 3850 � 20Ma range obtained by whole-rock 39Ar-40Arsystem (9, 10, 18) confirms that the argonsystem yields reliable age estimates. The agedifference may reflect the imprecisely known40K decay constant. A recalculation of theargon ages with the decay value proposed by(19) results in a �1% age increase.

We were able to constrain the SaU 169provenance even more tightly by using LunarOrbiter images in combination with Clemen-tine-derived Fe-Ti maps, Lunar ProspectorTh maps, Clementine color-ratio images(750/950 nm), lunar crater ages (20, 21) (ta-ble S7), and data from the Apollo and Lunamissions. The source location is based on thefollowing assumptions: (i) The high Th (32.7ppm) impact melt is derived from one of thehighest Th areas defined by Lunar Prospector�-ray mapping; (ii) The regolith Fe-Ti-Thconcentrations are representative of the rego-lith of the ejection area; (iii) The age distur-bances at �2800 and 200 Ma were caused byimpacts that were able to disturb the Ar sys-tem; (iv) The ejection crater from which therock was launched to Earth is on the order ofa few kilometers (17).

The Lunar Prospector �-ray mappingdemonstrated that the surface occurrence ofKREEP is confined to the near-side Procel-larum terrane (2–5, 22, 23). Because the bulkTh content of the impact melt (Table 1) ishigher than any pixel of the Lunar Prospectormeasurements (maximum, 9.36 to 10.40ppm) (7), we assume that SaU 169 must comefrom one of the Th hot spots (fig. S3), whichare located at the Aristarchus and Aristillus

craters, the Montes Carpathus–Fra Mauro re-gion, and the Lalande crater region (7).Among these Th hot spots, only the areaaround Lalande and southeast of the craterAristillus are compatible with our regolithFe-Ti-Th concentrations (table S7). In bothareas, a young (bright-halo), small crater as apossible launch-to-Earth crater is evident.However, the age data for SaU 169 render theLalande crater area (9°W 5°S) the most plau-sible because it contains a �2800 Ma crater(Lalande, 2246 to 2803 Ma) and a �200 Macrater (e.g., Lalande A, 175 to 300 Ma) (tableS7). Based on the comprehensive data setpresented, we conclude that an origin fromthe Lalande crater area is most likely.

SaU 169 is the only large high-KREEPsample for which simultaneous determina-tions of precise age and chemistry are avail-able. The Imbrium impact age dates a strongdecline of large impacts in the inner solarsystem corresponding to the beginning oflife-supporting conditions on Earth.

References and Notes1. A. E. Metzger, J. I. Trombka, L. E. Peterson, R. C. Reedy,

J. R. Arnold, Science 179, 800 (1973).2. L. A. Haskin, J. L. Gillis, R. L. Korotev, B. L. Jolliff, J.Geophys. Res. 105, 20403 (2000).

3. B. L. Jolliff, J. J. Gillis, L. A. Haskin, R. L. Korotev, M. A.Wieczorek, J. Geophys. Res. 105, 4197 (2000).

4. R. L. Korotev, J. Geophys. Res. 105, 4317 (2000).5. M. A. Wieczorek, R. J. Phillips, J. Geophys. Res. 105,

20417 (2000).6. G. H. Heiken, D. T. Vaniman, B. M. French, LunarSource Book: A User’s Guide to the Moon (CambridgeUniversity Press, Cambridge, 1991).

7. D. J. Lawrence et al., Geophys. Res. Lett. 26, 2681 (1999).8. D. E.Wilhelms,U.S. Geol. Survey Prof. Pap.1348, 1 (1987).9. B. A. Cohen, T. D. Swindle, D. A. Kring, Science 290,

1754 (2000).10. D. Stoffler, G. Ryder, Space Sci. Rev. 96, 9 (2001).11. S. Russell et al., Meteorit. Planet. Sci. 38, A189 (2003).12. U. Wiechert et al., Science 294, 345 (2001).13. B. L. Jolliff, Int. Geol. Rev. 40, 916 (1998).14. C. Hohenberg, K. Marti, F. Podosek, R. Reedy, J. Shirck,

Proc. Lun. Planet. Sci. Conf. 9th, 2311 (1978).15. O. Eugster, T. Michel, Geochim. Cosmochim. Acta 59,

177 (1995).16. C. Tuniz et al., Geophys. Res. Lett. 10, 804 (1983).17. P. H. Warren, Icarus 111, 338 (1994).18. G. B. Dalrymple, G. Ryder, J. Geophys. Res. 101,

26069 (1996).19. K. Min, R. Mundil, P. R. Renne, K. R. Ludwig, Geochim.

Cosmochim. Acta 64, 73 (2000).20. R. B. Baldwin, Icarus 71, 19 (1987).21. R. B. Baldwin, Icarus 61, 63 (1985).22. R. C. Elphic et al., J. Geophys. Res. 107, E4, 8-1 (2002).23. W. C. Feldman, O. Gasnault, S. Maurice, D. J. Lawrence,

R. C. Elphic, J. Geophys. Res. 107, E3, 5-1 (2002).24. Materials and methods are available as supporting

material on Science Online.25. Funded by Swiss National Foundation grants 21-

64929.01, 20-61933.00, and credit 21-26579.89.Thanks to H. Al-Azri and K. Musallam for support; toL. R. Gaddis, USGS, for Lunar Orbiter images; and toJ. Kramers, J. Ridley, L. Diamond, D. Fleitmann, T.Armbruster, and B. Hacker for reviews.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/657/DC1Materials and MethodsSOM TextFigs. S1 to S3Tables S1 to S7References


Fig. 3. 207Pb/206Pb ages of 12 zircons from theimpact melt. Data are corrected for common Pb(24) and yield a weighted average age of3909 � 13 Ma (2�; MSWD � 0.33).

R E P O R T S


Foundering Lithosphere ImagedBeneath the Southern

Sierra Nevada, California, USAOliver S. Boyd,* Craig H. Jones, Anne F. Sheehan

Seismic tomography reveals garnet-rich crust and mantle lithosphere descend-ing into the upper mantle beneath the southeastern Sierra Nevada. The de-scending lithosphere consists of two layers: an iron-rich eclogite above amagnesium-rich garnet peridotite. These results place descending eclogiteabove and east of high P wave speed material previously imaged beneath thesouthern Great Valley, suggesting a previously unsuspected coherence in thelithospheric removal process.

The Sierra Nevada mountain range of Cali-fornia is one of the highest (about 3 km meanelevation) in the United States; however, thecrust is only 35 km thick (1) and requiressome unusual structure in the mantle. Xeno-liths (2) and volcanic rocks (3, 4) suggest thatthe upper mantle source for these materials,beneath the Sierra Nevada, changed from aneclogite facies garnet pyroxenite to a garnet-free spinel peridotite about 4 million yearsago (Ma). Eclogitic rocks may still be presentin the upper mantle below the Sierra Nevada(5), and knowing where and how much is leftconstrains how lithospheric material is re-moved and how continental crust grows andchanges beneath mountain ranges (6). To de-termine whether there are any eclogitic rocksleft, we used three-dimensional models of Pand S wave speeds to define compositionaland thermal characteristics of the lithosphere.We measured attenuation to correct the ve-locities for anelastic effects, which includetemperature and hydration. The correctedwave speeds are subsequently termed anhar-monic wave speeds and compared with lab-oratory predictions.

Low Pn velocities directly beneath the highSierra (7, 8) and high attenuation in the uppermantle in the region (9) are consistent with hightemperature, low-density material. Shear wavesplitting measurements determined from theseismic phase SKS (10, 11) (Fig. 1) indicatestrong seismic anisotropy oriented N80°E be-neath the central Sierra, probably because of thepresence of strained, olivine-rich, peridotiticmantle. Less splitting beneath the eastern andwestern Sierra may indicate largely isotropicmaterial, such as eclogite (12), little strain, or avertical fast axis of anisotropy. The region oflarge SKS splitting and low Pn velocity corre-lates well with high mantle electrical conduc-

tivity that is probably due to 1% partial meltbeneath the southern Sierra Nevada (13).These observations are consistent with thepresence of a peridotitic uppermost mantlenear asthenospheric temperatures near theSierran crest today.

We recorded 40 teleseismic events thatyielded 800 seismic wave traces on 24 broad-band seismometers (Fig. 1). We measured

direct P wave and Sfast and Sslow wave traveltime delays (tP, tSf

, and tSs) and Sfast and Sslow

wave path–integrated attenuation, �t* (14),to invert for variations in P wave speed, vP,fast and slow S wave speeds (vSf

and vSs), and

fast- and slow-oriented S wave attenuation(QSf

-1 and QSs

-1). These were combined to ex-amine variations in the ratio of P wave to Swave velocity, vP /vS, and transverse S waveanisotropy. Our tomographic inversions re-duce the variance of the data by more thanhalf of that available, whereas measurementnoise and station and event statics account forthe remainder (15).

The tomographic models reveal regions ofalternating high and low velocity and attenu-ation that dip �55° to the east southeast (Fig.2). The region of high P wave velocity at36.4°N extends from –119.5°W to –118°Wand may be what has been referred to as theIsabella anomaly. The volume above thisanomaly consists of a slow P and S waveregion. The S wave velocities decrease moreacross this contrast than do the P wave ve-locities, giving rise to an increase in the vP /vS

ratio. These seismic anomalies trend north-east (Fig. 3C).

Department of Geological Sciences, University of Col-orado at Boulder, 2200 Colorado Avenue, Boulder, CO80309–0399, USA.


Fig. 1. Topographic map of the southern Sierra Nevada overlain with tomographically predictedshear wave splitting (this study) and independently determined SKS splitting measurements, filledcircles with white (10) or black (11) outline. Map of Nevada and California with pink outline ofstudy area given in the upper right for reference. The SKS measurements are placed at 125-kmdepth along the SKS ray path. The tomographically predicted splitting time is determined byintegrating differences between the fast and slow S wave speed models in the upper 200 km forregions with greater than 150-km thickness of resolution greater than 0.5. Thick purple line crossingat 36.4°N is the latitudinal position of the vertical slices in Fig. 2. The plus symbols are seismographstations used for the tomography, the multiplication symbols are stations used to measure SKSsplitting, and the asterisks, plus symbols overlain by multiplication symbols, are stations used forboth tomography and SKS splitting. The open boxes are garnet peridotite and garnet pyroxenitexenolith localities; solid boxes are spinel peridotite xenolith localities.

R E P O R T S


Upon vertically integrating the differencesin Sfast and Sslow travel times for the top 200km, we are able explain more than 50% of thevariance of the SKS splitting measurements(Fig. 1 and fig. S1) (15), an independentverification of our subsurface distribution inanisotropy. The SKS measurements were notused in the inversion; we used only the ori-entation of fast S wave speed derived fromthe splitting analysis. The shallow depth forthe origin of the splitting (�200 km) is con-sistent with the 1-s variation in splitting am-plitudes over short distances (�50 km) andinconsistent with splitting distributed overgreater depths.

Attenuation is needed to distinguish be-tween the effects on wave speed from ther-mal and compositional variations (16, 17).Anisotropy (12) and Poisson’s ratio (18)can then be used to constrain the composi-

tion (Table 1). Because of anelasticity, tem-perature variations will cause seismic ve-locities and attenuation to vary predictably(19). We interpret deviations from the ther-mal relationship as being caused by varia-tions in composition.

Because the dipping layer of low wavespeeds is accompanied by low attenuation,temperatures are relatively low (�200 Klower than the material to the east), andpartial melt or partially water-saturatedminerals are not likely present. This leavessome sort of compositional effect produc-ing, in addition to low velocities, a highvP/vS ratio and variable anisotropy. HighvP/vS ratios can indicate an increase in theamount of pyroxene and garnet relative toolivine or a decrease in the Mg number[Mg/(Mg � Fe)]. The variable anisotropyis more difficult to interpret. Lower

anisotropies can be from an absence ofolivine, an absence of strain, or a heteroge-neously strained peridotite. In response toinfilling asthenosphere, the low anisotropyin the inferred region of spinel peridotite(Fig. 2D) could be anisotropic but have anear-vertical axis of orientation. Normallyisotropic lithologies like eclogites might beanisotropic if pervasively cut by dikes or ifthe pyroxenes have become preferentiallyaligned (15). The large anisotropy appear-ing below the dipping layer of high P wavevelocities may reflect lateral extrusion ofasthenospheric peridotite in response todelaminating mantle lithosphere.

Ambiguity in interpreting the tomo-graphic results can be reduced by examin-ing the xenoliths erupted nearby in thecentral Sierra Nevada. There are threegroups of xenoliths: garnet pyroxenites andgarnet peridotites (both erupted before 8Ma) constituting the old mantle lithosphere(2, 20) and fertile, olivine-rich, garnet-freespinel peridotites (erupted since 4 Ma) withan asthenospheric affinity (2). Chemically,the garnet and pyroxene in the pyroxenitestend to have low Mg, whereas the olivine inthe peridotites has a higher Mg number (2).This observation means that the garnet py-roxenite will have higher vP/vS ratios rela-tive to the peridotites.

We calculated the P and S wave speeds(15, 21) for a range of expected compositionsand compare these values to our observationsof anharmonic wave speed (15) to broadlydetermine the mineralogy and Mg numbers inthe observed seismic anomalies (Fig. 3). Themodal proportions for the garnet pyroxenitexenolith samples are predicted to be about65% clinopyroxene and 35% garnet (22). TheMg number ranges from 0.6 to 0.9 for theclinopyroxenes and 0.3 to 0.6 for the garnet(2). The composition that best matches ourseismic observations averages 70% pyroxeneand 30% garnet (15). Compositions having asmuch as 40% garnet would be acceptable if aReuss average, as opposed to the Voight-Reuss-Hill average (23) used here, is favored.One garnet peridotite sample has 75% oli-vine, 15% orthopyroxene, 5% clinopyroxene,and 5% garnet (24), whereas the Mg numbersare reported to be on average 0.91, 0.91, 0.92,and 0.85 for those minerals, respectively (2).Our seismically determined composition av-erages 75% olivine, 20% pyroxene, and 5%garnet. The young peridotite samples are devoidof garnet and average 80% olivine, with the re-mainder orthopyroxene, clinopyroxene, and spinel(25). Their Mg numbers average 0.89, 0.89, 0.87,and 0.67, respectively (2). Because a majority ofthe young samples have been classified as har-zburgites, we restrict our analysis to clinopyrox-ene-free assemblages. We find this compositionhas on average 80% olivine, 16% pyroxene, and4% garnet (26).

Fig. 2. Vertical slices of our tomographic models and derived quantities:percent change in P waveslowness (A), change in attenuation (B), percent change in anharmonic vP/vS ratio where vS is takenfrom the average of the fast and slow models (C), and shear wave anisotropy where the slow modelis derived as residuals from the fast model (D). The solid lines indicate regions of descending garnetperidotite, and the dashed lines delineate regions of garnet pyroxenite. The regions of low velocitiesand high attenuation above the garnet pyroxenite are presumably the infilling of asthenosphericspinel peridotite (dashed-dot outlines). Topography is depicted with 10 times vertical exaggeration.

Table 1. Changes observed in geophysical properties in the mantle from different causes. The asteriskindicates that the attenuation will only increase if high temperatures are required to produce melt (15).The question mark indicates that anisotropy may increase or change orientation if melt pockets and/orbands are favorably oriented.

Increasing factor vP vP/vS Attenuation Anisotropy Density

Temperature Decrease Increase Increase No change DecreaseMelt Decrease Increase * ? DecreaseMagnesium Increase Decrease No change No change DecreaseGarnet/olivine ratio Increase Increase No change Decrease Increase

R E P O R T S


Many eclogites are seismically fast be-cause they have a large proportion of garnetand much of the pyroxene is in the form ofjadeite (27). The xenoliths found here, how-ever, have large amounts of clinopyroxene(24) and very little jadeite (2). The garnetpyroxenites are therefore slower than com-mon eclogites and have a higher vP /vS ratiorelative to the peridotites (Fig. 3A). A com-parison of the characteristics of the seismicanomalies to calculated values for mineralassemblages (Fig. 3A) reveals that the dip-ping layer of high velocity is the garnet pe-ridotite; the low velocity layer above, garnetpyroxenite; and the seismically intermediateregion above and to the east, asthenosphericspinel peridotite (Fig. 2). A map view of theolder garnet-bearing assemblages at 100-kmdepth indicates a northeast strike to this struc-ture (Fig. 3B). The seismically determinedcompositions imply that, at 100-km depth,densities increase from close to neutrallybuoyant garnet peridotite below to verydense, negatively buoyant (�� 185 kg/m3)garnet pyroxenite above.

Although we have suggested specific dif-ferences in composition and temperature forthe three rock assemblages identified, themost robust conclusion is that three distinctrock types exist in the upper mantle beneaththe Sierra Nevada. This observation is madepossible by correcting measured P and Swave seismic velocities for attenuation andcomparing these values to laboratory results.The apparent dip of the two garnet-bearingpackages and the asymmetry of the package(garnet peridotite to the west of the garnetpyroxenite) are important constraints on themechanism of removal of this material. The

asymmetry resembles delamination of strati-fied mantle lithosphere from the crust and isinconsistent with most numerical models ofremoval by development of a Rayleigh-Taylor instability (28), although such asym-metry might be produced by an initiallyasymmetric instability (29). The eastwardplunge of this body indicates that either ma-terial has moved directly downward or some-what to the east relative to the overlyingcrust, in contrast to motion to the west rela-tive to the crust inferred from older images ofthis structure (5). Imaging one of the very few(if not the only) examples of ongoing remov-al of mantle lithosphere from beneath conti-nental crust, this tomography provides someof the best observational constraints on theprocess of removing dense material fromcontinental crust.

References and Notes1. B. Wernicke et al., Science 271, 190 (1996).2. M. Ducea, J. Saleeby, J. Geophys. Res. 101, 8229 (1996).3. C. R. Manley, A. F. Glazner, G. L. Farmer, Geology 28,

811 (2000).4. G. L. Farmer, A. F. Glazner, C. R. Manley, Geol. Soc.

Am. Bull. 114, 754 (2002).5. G. Zandt, Int. Geol. Rev. 45, 213 (2003).6. M. Ducea, J. Geophys. Res. 107, 10.1029/

2001JB000643 (2002).7. C. H. Jones, H. Kanamori, S. W. Roecker, J. Geophys.

Res. 99, 4567 (1994).8. M. M. Fliedner et al., Geology 24, 367 (1996).9. H. H. Al-Khatib, B. J. Mitchell, J. Geophys. Res. 96,

18129 (1991).10. J. Polet, H. Kanamori, Geophys. J. Int. 149, 313 (2002).11. C. H. Jones, R. A. Phinney, Geol. Soc. Am. Abstr. Prog.

31 (no. 7), A-481 (1999).12. N. I. Christensen, W. D. Mooney, J. Geophys. Res.

100, 9761 (1995).13. M. Ducea, S. K. Park, Geophys. Res. Lett. 27, 2405

(2000).14. A. F. Sheehan, S. C. Solomon, J. Geophys. Res. 97,

15339 (1992).15. Please refer to the supporting online material on Sci-

ence Online for more information about seismic meth-ods, resolution, and calculation of mineral properties.

16. O. S. Boyd, A. F. Sheehan, in Lithospheric Structureand Evolution of the Rocky Mountain Region, K. E.Karlstrom, G. R. Keller, Eds. (Geophys. Monogr. Am.Geophys. Union), in press.

17. S. Karato, Geophys. Res. Lett. 20, 1623 (1993).18. N. I. Christensen, J. Geophys. Res. 101, 3139 (1996).19. I. Jackson, Annu. Rev. Earth Planet. Sci. 21, 375

(1993).20. F. C. W. Dodge, J. P. Lockwood, L. C. Calk, Geol. Soc.

Am. Bull. 100, 938 (1988).21. B. R. Hacker, G. A. Abers, Geochem. Geophys. Geosys.

5, 10.1029/2003GC000614 (2004).22. J. Saleeby, M. Ducea, D. Clemens-Knott, Tectonics 22,

10.1029/2002TC001374 (2003).23. J. P. Watt, G. F. Davies, J. O’Connell, Rev. Geophys.

Space Phys. 14, 541 (1976).24. M. Ducea, J. Saleeby, Contrib. Mineral. Petrol. 133,

169 (1998).25. H. G. Wilshire et al., U.S. Geol. Surv. Prof. Pap. 1443

(1988), p. 179.26. Garnet is used for the spinel peridotite composition

because we are considering seismic velocities at 3GPa, by which pressure spinel would have trans-formed to garnet.

27. R. G. Coleman, D. E. Lee, J. B. Beatty, W. W. Brannock,Geol. Soc. Am. Bull. 76, 483 (1965).

28. G. A. Houseman, P. Molnar, Geophys. J. Int. 128, 125(1997).

29. M. Jull, P. B. Kelemen, J. Geophys. Res. 106, 6423(2001).

30. Thanks to P. Molnar and L. Farmer for helpful discus-sion and B. Phinney and several land use agencies(U.S. Forest Service, Sequoia and Kings Canyon Na-tional Parks, and Bureau of Land Management) forhelp in obtaining the data. We also acknowledge twoanonymous reviewers and L. Rowan for thoroughreviews. This work was supported by NSF grants9526974 and 0003747 and a Cooperative Institutefor Research in Environmental Research Sciencesfellowship.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/660/DC1Materials and MethodsFigs. S1 to S3Tables S1 and S2


Fig. 3. (A) Percentchange in anharmonicvP/vS ratio versus per-cent change in anhar-monic P wave slow-ness for a range ofmodel compositionsat a pressure of 3 GPaand a range of tem-peratures. Composi-tions defined in thetext that best matchthe seismic observa-tions are garnet py-roxenite (red dia-mond) at 1200°C,spinel peridotite (redcircle) at 2000°C, andgarnet peridotite (redtriangle) at 1000°C.Thick black lines rep-resent ranges in composition where garnet is substituted for pyroxene.For the garnet pyroxenite, garnet varies from 20% (right end of black line,slower wave speeds) to 40%; the spinel peridotite, 0 to 10%; and thegarnet peridotite, 0 to 10%. Yellow lines represent a 400 K variation intemperature (900 to 1300°C), with faster wave speeds (to the left)accompanying lower temperatures. The blue line spans the range of Mgnumbers for the garnet pyroxenite samples. Contours denote anharmonicP wave slowness and vP/vS from our Sierran tomography from 75- to

100-km depth. Each contour interval represents 20% of the data. Col-ored areas correspond to compositional regions outlined in the to-mography of Fig. 2 and shaded areas in the 100-km-depth slice (B).Garnet pyroxenite, green; spinel peridotite, red; and garnet peridotite,blue. Compositional zones in (B) derived from tomography usingcomposition relations (15) with P wave slowness, Poisson’s ratio,anisotropy, and attenuation. The white dashed line delimits the partof the tomography model with resolution greater than 0.4.

R E P O R T S


Herbivores Promote HabitatSpecialization by Trees in

Amazonian ForestsPaul V. A. Fine,1,2* Italo Mesones,3 Phyllis D. Coley1

In an edaphically heterogeneous area in the Peruvian Amazon, clay soils andnutrient-poor white sands each harbor distinctive plant communities. To de-termine whether a trade-off between growth and antiherbivore defense en-forces habitat specialization on these two soil types, we conducted a reciprocaltransplant study of seedlings of 20 species from six genera of phylogeneticallyindependent pairs of edaphic specialist trees and manipulated the presence ofherbivores. Clay specialist species grew significantly faster than white-sandspecialists in both soil types when protected from herbivores. However, whenunprotected, white-sand specialists dominated in white-sand forests and clayspecialists dominated in clay forests. Therefore, habitat specialization in thissystem results from an interaction of herbivore pressure with soil type.

The floras of tropical forests in areas ofhigh environmental heterogeneity havehigh beta diversity, leading many authors toconclude that habitat specialization in trop-ical plants contributes significantly to theglobal diversity gradient (1–4 ). The changein species composition between habitats(beta diversity) in plants is thought to becaused primarily by physiological adapta-

tions to different abiotic conditions (5–7 ).But herbivores have been shown to affectplant population dynamics in many ways,including plant distributions (8–10). Thebest way to test the effect of herbivores onplant distributions is with transplant exper-iments involving multiple species, but fewsuch studies have been conducted. Al-though they have evaluated only one or twospecies at a time, these studies have con-sistently shown that herbivores can signif-icantly limit plant distributions, often re-stricting species to a subset of the habitatsthey could physiologically tolerate in theabsence of herbivores (11–15). However,no such studies have been conducted intropical forests [but see (16, 17 )], whereherbivore abundance and rates of herbivory

far exceed those of the temperate zone (18).If herbivores interact with abiotic grad-ients to intensify differences in habitatquality, this phenomenon could account forthe high degree of habitat specialization inenvironmentally heterogeneous areas with-in tropical systems.

The lowland Amazonian ever-wet rain-forest in the Allpahuayo-Mishana Reservenear Iquitos, Peru (3°57�S, 73°26�W), pro-vides an ideal system to study habitat spe-cialization and the role of insect herbivores.Forests in the Iquitos area occur on a mo-saic of soil types with well-defined bound-aries, including extremely infertile white-sand soils adjacent to lateritic red-clay soils(19). These two soil types are each charac-terized by a highly distinctive but relatedflora (19–22). Many white-sand specialisttrees belong to the same genera as neigh-boring clay forest specialists, allowing for aphylogenetically controlled experiment us-ing edaphic specialist species.

Janzen (23) argued that the most criticalfactors in the evolution of white-sand special-ist plants were herbivores and plant defensesrather than physiological tolerance to poorsoils. A mechanism that would produce sucha result is a trade-off between allocations togrowth and defense that selects for differenttraits in white-sand and clay forest species.We hypothesized that plants that do not in-vest sufficiently in defenses ought to be ex-cluded from white-sand forests by herbi-vores, because the cost of replacing damagedtissue in such a nutrient-poor environmentshould be prohibitively high (24). In contrast,species that invest heavily in defenses shouldgrow more slowly and be at a competitivedisadvantage in clay forests.

1Department of Biology, University of Utah, 257 S.1400 East, Salt Lake City, UT 84112, USA. 2Environ-mental and Conservation Programs and Departmentof Botany, Field Museum of Natural History, 1400 S.Lake Shore Drive, Chicago, IL 60605, USA. 3Depart-ment of Forestry, Universidad Nacional de la Amazo-nıa Peruana, Plaza Serafın Filomeno 246, Iquitos, Peru.


Fig. 1. Phylogenetic re-lationships of the spe-cies in the transplantexperiment with theirfamily, genus, species,and soil type (origin).The phylogenetic treeis adapted from (37),and the Burseraceaephylogeny is adaptedfrom (38). Tetragastrispanamensis has beenfound to be imbeddedwithin Protium sensulato (38), therefore weare considering Tetra-gastris panamensis as aProtium clay specialistspecies. Soil type (ori-gin) was determinedby extensive tree andseedling inventoriesthat we conducted inmore than 25 sitesthroughout the 57,600-ha Allpahuayo-Mishana Reserve. Our designationof white-sand specialists is consistent with other inventories and a publishedflora from the area (19, 22, 38, 39). Our designation of clay specialists

conforms to other species lists on clay soils in the area (19, 22, 38) as wellas with lists from the Ecuadorian Amazon, which is almost entirely com-posed of clay soils and where white-sand forests are unknown (40).

R E P O R T S


Using a 21-month reciprocal transplantexperiment with 20 common white-sandandclay forest species from six genera, andmanipulating the presence of herbivores,we directly assessed the effect of herbivoryon the maintenance of habitat specializa-tion in trees. We addressed three mainquestions: (i) Can soil specialists grow in adifferent edaphic environment, or is soiltype alone a sufficient barrier for theseplants? (ii) Do herbivores exclude clay spe-cialists from white-sand forests? (iii) Dowhite-sand specialists grow more slowlythan clay forest specialists, putting them ata competitive disadvantage in clay forests?

We transplanted 880 seedlings from sixgenera arranged in pairs of clay/white-sandspecialist congeners (Fig. 1). In May 2001,we built 22 control and 22 herbivore exclo-sures (3 m by 3 m by 2 m) in clay andwhite-sand habitats, controlling for light lev-

els (25). One seedling of each species wastransplanted into each of the 22 controls and22 exclosures, for a total of 20 seedlings perexclosure. The exclosures were completelycovered by 1-mm fine nylon netting, whereasthe controls had netting only on the roof. Theexperiment lasted until February 2003. Toassess differences between seedlings inwhite-sand versus clay forests, and in controlversus protected treatments (exclosures), wemeasured growth rate (leaf area per day andmeristem height per day) and mortality (25).

Both clay and white-sand specialist spe-cies exhibited positive growth on both soiltypes. In white-sand habitat, clay specialistsand white-sand specialists suffered statistical-ly similar mortality when protected from her-bivores (Fig. 2). These findings contradict thehypothesis that white-sand soil is probablytoxic to nonadapted edaphic specialists be-cause of differences in H�/Al ratios (26).Our results also contradict a related hypoth-esis that posits that the low nutrient availabil-ity and/or water retention capacity of whitesand exclude most nonspecialist plants fromwhite-sand forests (6, 27, 28). Our resultsshowed that when protected from herbivoresand planted in white-sand habitat, clay spe-cies produced more than twice the amount ofleaf area [0.21 cm2 day�1 versus 0.09 cm2

day�1, P � 0.05 (Fig. 2A)] and grew signif-icantly taller than white-sand species [0.09mm day�1 versus 0.05 mm day�1, P � 0.05(Fig. 2B)]. Summing the 18 months of mea-surements, protected clay specialists grew anaverage of 4.9 cm in height and 115.3 cm2 inleaf area as compared to protected white-sandspecialists’ 2.7 cm in height and 49.4 cm2 inleaf area. Although soil type does have asignificant effect on seedling meristemgrowth rate (P � 0.05, table S3), clearly therestriction of clay specialist species to clayhabitat cannot be explained solely as a func-tion of soil type, at least not at the criticalseedling stage.

Our results showed a strong effect of her-bivores on the exclusion of clay specialistspecies from white-sand forests. When un-protected from herbivores in white-sandhabitat, white-sand specialists exhibited asignificantly higher leaf growth rate than clayspecialists [0.15 cm2 day�1 versus –0.4 cm2

day�1, P � 0.05 (Fig. 2A)] and suffered onlyhalf the mortality rate [9% versus 20%, P �0.05 (Fig. 2C and fig. S2)]. Taken together,the results from white-sand forests indicatethat without the presence of herbivores, clayspecialists survived just as well as white-sandspecialists but grew taller and produced moreleaf area and thus potentially could invadeand displace white-sand specialists fromwhite-sand forests.

Furthermore, white-sand specialist andclay specialist species responded differentlyto herbivore protection in both leaf and mer-

istem growth rate [origin � protection inter-action, P � 0.0001 and P � 0.01, respective-ly (tables S2 and S3)] and in mortality rate(P � 0.05) (Fig. 2). White-sand species didnot respond to protection in either habitat (P �0.05), presumably because their leaves arealready highly defended against herbivores(23, 29). Indeed, chemical analyses of theseseedlings showed that leaves of white-sandspecialists contain significantly higher tannin:protein ratios than leaves of clay specialists(30). In contrast, clay specialist species gained asignificant advantage from herbivore protec-tion, producing 267% more leaf area per daywhen protected (Fig. 2A).

All species grew better on clay soils, butclay specialists outperformed white-sand spe-cialists on clay soils, providing indirect evi-dence that interspecific competition limits theinvasion of clay forests by white-sand spe-cies. In the clay soil, white-sand species ex-hibited significantly slower height (P � 0.01)and leaf (P � 0.01) growth than clay special-ists [habitat � origin interaction (Fig. 2, Aand B, and tables S2 and S3)]. These resultsare consistent with the hypothesis that there isa growth/defense trade-off in these tree seed-lings and that the high defense levels ofwhite-sand specialists preclude a rapidgrowth rate and place them at a competitivedisadvantage in clay soils.

We conclude that herbivores play an impor-tant role in maintaining habitat specialization inclay and white-sand specialist trees. Althoughother factors certainly influence habitat spe-cialization at other stages of a tree’s life (suchas seed predation, the ability to tolerate fall-ing debris, root morphology, etc.), herbivoresappear to have primary importance at theseedling stage, and in addition, likely contin-ue to attack trees throughout a tree’s life.Herbivores killed a significant percentage oftree seedlings in this system, and in white-sand forests they selectively attacked clayspecialist transplants, maintaining the domi-nance of endemic white-sand specialists.Without herbivores, our results indicate thatclay specialists outperform white-sand spe-cialists in both soil types. However, withherbivores present, clay specialist speciesdominate only in clay forests and white-sandspecialists dominate only in white-sand for-ests. The strong negative impact of herbi-vores on the growth and mortality rates ofclay specialists underscores the intense selec-tive pressure in white-sand forests favoring alarge investment in antiherbivore defenses.Thus, the trade-off between growth rate andantiherbivore defense appears to restrict spe-cies to one habitat or the other for the sixgenera we studied, and habitat specializationresults from an interaction of herbivore pres-sure with soil type.

Edaphic heterogeneity, especially in thetropics, has been implicated in the generation

Fig. 2. The effects of habitat and herbivore pro-tection on (A) leaf area growth rate, (B) meristemheight growth rate, and (C) percent mortality forwhite-sand and clay specialist species. Bars rep-resent mean and �1 SE. Values with differentletters (a, b, and c) are significantly different fromone another [Tukey tests for (A) and (B); Mann-Whitney U for (C)].

R E P O R T S


and maintenance of high diversity (1–4, 20,31, 32), even though the specific mechanismsby which edaphic factors might influencediversity have, until now, never been tested(33). Our results suggest a mechanism bywhich herbivores may influence plant evolu-tion: by accentuating habitat differences andthereby increasing the potential for edaphicheterogeneity to produce habitat special-ization. There is mounting evidence thatparapatric speciation across environmentalgradients can occur in the face of gene flow(34–36). Thus, our data suggest that herbi-vores can be viewed as a diversifying forcebecause they make existing abiotic gradientsmore divergent and cause finer-scale habitatspecialization by magnifying the differencesbetween habitats. Our study highlights the im-portance of edaphic differences, but the samemechanism could work across other abiotic gra-dients such as altitude, rainfall, and floodingregime—wherever the impact of herbivores isdissimilar across a habitat boundary.

References and Notes1. A. H. Gentry, Ann. MO Bot. Gard. 75, 1 (1988).2. D. B. Clark, M. W. Palmer, D. A. Clark, Ecology 80,

2662 (1999).3. H. Tuomisto et al., Science 269, 63 (1995).4. H. Tuomisto, K. Ruokolainen, M. Yli-Halli, Science

299, 241 (2003).5. J. L. Harper, The Population Biology of Plants (Aca-

demic Press, New York, 1977).6. D. A. Coomes, P. J. Grubb, Vegetatio 122, 167 (1996).7. A. Mark, J. Dickinson, J. Allen, R. Smith, C. West,

Austral Ecol. 26, 423 (2001).8. S. Louda, K. H. Keeler, R. D. Holt, in Perspectives in

Plant Competition, J. Grace, D. Tilman, Eds. (AcademicPress, New York, 1990), pp. 413–444.

9. N. S. Cobb et al., Oecologia 109, 389 (1997).10. H. Olff, M. Ritchie, Trends Ecol. Evol. 13, 261 (1998).11. M. Parker, R. Root, Ecology 62, 1390 (1981).12. S. M. Louda, Ecol. Monogr. 52, 25 (1982).13. S. M. Louda, Ecology 64, 511 (1983).14. S. M. Louda, J. Rodman, J. Ecol. 84, 229 (1996).15. C. D. G. Harley, Ecology 84, 1477 (2003).16. S. J. DeWalt, J. S. Denslow, K. Ickes, Ecology 85, 471

(2004).17. R. T. King, Biotropica 35, 462 (2003).18. P. D. Coley, J. A. Barone, Annu. Rev. Ecol. Syst. 27, 305

(1996).19. K. Ruokolainen, H. Tuomisto, in Geoecologia y Desar-

rollo Amazonico: Estudio Integrado en la Zona deIquitos, Peru, R. Kalliola, S. Flores, Eds. (Univ. of TurkuPress, Turku, Finland, 1998), pp. 253–368.

20. A. H. Gentry, in Conservation Biology: The Science ofScarcity and Diversity, M. Soule, Ed. (Sinauer, Sunder-land, MA, 1986), pp. 153–181.

21. K. Ruokolainen, A. Linna, H. Tuomisto, J. Trop. Ecol.13, 243 (1997).

22. R. Vasquez M., Florula de las Reservas Biologicas deIquitos, Peru (Missouri Botanical Garden Press, St.Louis, MO, 1997).

23. D. H. Janzen, Biotropica 6, 69 (1974).24. P. D. Coley, J. P. Bryant, F. S. Chapin, Science 230, 895

(1985).25. Detailed information on materials and methods is

available on Science Online.26. J. Proctor, Bot. J. Scotl. 51, 1 (1999).27. E. Medina, E. Cuevas, in Mineral Nutrients in Tropical

Forest and Savanna Ecosystems, J. Proctor, Ed. (Black-well, Oxford, 1989), pp. 217–240.

28. J. A. Moran, M. G. Barker, A. J. Moran, P. Becker, S. M.Ross, Biotropica 32, 2 (2000).

29. D. McKey, P. G. Waterman, N. Mbi, J. S. Gartlan, T. T.Struhsaker, Science 202, 61 (1978).

30. P. V. A. Fine et al., unpublished data.

31. K. Young, B. Leon, Brittonia 41, 388 (1989).32. R. Condit et al., Science 299, 666 (2002).33. P. Sollins, Ecology 79, 23 (1998).34. J. A. Endler, Geographic Variation, Speciation, and

Clines (Princeton Univ. Press, Princeton, NJ, 1977).35. C. Moritz, J. Patton, C. Schneider, T. B. Smith, Annu.

Rev. Ecol. Syst. 31, 533 (2000).36. R. Ogden, R. S. Thorpe, Proc. Natl. Acad. Sci. U.S.A.

99, 13612 (2002).37. M. W. Chase et al., Ann. MO Bot. Gard. 80, 528 (1993).38. P. V. A. Fine, D. C. Daly, G. Villa, I. Mesones, K. M.

Cameron, unpublished data (available on requestfrom the first author).

39. R. Garcıa-Villacorta, M. Ahuite R., M. Olortegui Z.,Folia Amazonica 14, 11 (2002).

40. P. Jørgensen, S. Leon-Yanez, Catalogue of the Vascu-lar Plants of Ecuador (Missouri Botanical GardenPress, St. Louis, MO, 1999).

41. We thank the Peruvian Ministry of Natural Resources(INRENA) for permission to conduct this study; D. DelCastillo, L. Campos, E. Rengifo, and S. Tello of theInstituto de Investigaciones de la Amazonıa Peruana

(IIAP) for logistical support and permission to work inand around the Estacion Allpahuayo; E. Aquituari, M.Ahuite, J. Guevara, M. Jackson, M. Olortegui, J. Reed,and F. Vacalla for field assistance; J. Alvarez, L. Bohs,D. Dearing, D. Feener, R. Foster, T. Kursar, S. Schnitzer,J. Kircher, and H. Stevens for advice; and S. DeWalt, S.Louda, N. Pitman, H. Howe, N. Cordeiro, M. Jorge, G.Nunez, P. Sethi, A. Sullivan, B. Zorn-Arnold, and J.Lokvam for helpful comments regarding the manu-script. Funding was provided by an NSF PredoctoralFellowship to P.V.A.F. and a NSF Doctoral Disserta-tion Improvement Grant to P.V.A.F. and P.D.C.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/663/DC1Abstract in SpanishMaterials and MethodsFigs. S1 and S2Tables S1 to S3References


Silencing the Jasmonate Cascade:Induced Plant Defenses and

Insect PopulationsAndre Kessler, Rayko Halitschke, Ian T. Baldwin*

We transformed the native tobacco, Nicotiana attenuata, to silence its lipoxy-genase, hydroperoxide lyase, and allene oxide synthase genes in order to inhibitoxylipin signaling, known to mediate the plant’s direct and indirect defenses.When planted into native habitats, lipoxygenase-deficient plants were morevulnerable to N. attenuata’s adapted herbivores but also attracted novel her-bivore species, which fed and reproduced successfully. In addition to highlight-ing the value of genetically silencing plants to study ecological interactions innature, these results show that lipoxygenase-dependent signaling determineshost selection for opportunistic herbivores and that induced defenses influenceherbivore community composition.

The plant traits that are important for resis-tance to herbivore attack in nature are com-plex and operate on many spatial scales. Theyinvolve direct defenses (toxins, digestibilityreducers, etc.) (1), which themselves protectplants, as well as indirect defenses, whichwork with components of a plant’s commu-nity (natural enemies, diseases, etc.) (2–6).Moreover, plant defenses can be constitutive-ly expressed or produced in response to anattacking pathogen or herbivore, when theyare needed (7, 8).

Phenotypic plastic responses such as her-bivore-induced plant defenses are thought tobe an adaptation to unpredictable environ-ments (8). In native populations of Nicotianaattenuata in the southwestern United States,herbivory is an unpredictable selective factor.N. attenuata’s ephemeral occurrence afterfires (9) forces its herbivore community to

reestablish itself with every new plant popu-lation, and the plant produces a wide array ofdirect and indirect chemical defenses in re-sponse to this unpredictable herbivore attack(10). Many of the responses are specificallyelicited by signals introduced into woundsduring feeding (11), and most herbivore-induced responses studied to date requireoxylipin signaling (12, 13).

Genetic transformation provides a valu-able tool with which to manipulate traits thatmediate complex plant-herbivore interac-tions and allows an integrative analysis ofsingle traits (14–16 ). However, transfor-mants are usually evaluated with knownchallenges, not the vast number of un-knowns that occur in nature. We used trans-formed lines of the wild tobacco species N.attenuata, which express N. attenuata li-poxygenase 3 (NaLOX3), hydroperoxidelyase (NaHPL), and allene oxide synthase(NaAOS) in an antisense orientation (as-lox, as-hpl, as-aos, respectively) (17 ) tostudy herbivore-induced plant responses innature. All three enzymes are key regula-tors in two distinct oxylipin pathways (fig.

Department of Molecular Ecology, Max-Planck-Institute for Chemical Ecology, Hans-Knoll-Strasse 8,Jena 07745, Germany.


R E P O R T S


S1A) and are known to play a major role inplants’ wound recognition and signaling(13), but their defensive function in therough-and-tumble of the natural environ-ment is unknown.

In laboratory studies, plants deficient in theexpression or recognition of jasmonates (JAs)derived from LOX3 are unable to elicit defensecompounds and are more susceptible to herbi-vore attack (15, 18, 19). Similarly, silencingAOS in N. attenuata plants partially reduces JAand defense compound accumulation but doesnot attenuate the resistance to herbivores (20).HPL-derived C6-aldehydes and -alcohols(green leaf volatiles, or GLVs) are antimicrobi-al and can function as direct defenses againstsome herbivores (21). Moreover, some C6-compounds may function as indirect defenses(5) or play a role in eliciting defense geneexpression (22) and signaling within or be-tween plants (23). However, the function ofoxylipins is not restricted to wound signalingand defense induction but includes the regula-tion of tuber growth (15), trichome (19) andflower development (24, 25), and ultraviolet Bprotection. Thus, how LOX3-, AOS-, and HPL-deficient N. attenuata plants would behave intheir natural environment with its complexity ofstressors was unknown. To extend the labora-tory characterization of these genetically trans-formed plant lines (18, 20), we examined theirgrowth and induced production of volatile or-ganic compounds (VOCs), and evaluated theirresistance to a specialized herbivore (Manducasexta) under field conditions. We transplantedyoung rosette plants into the field plots andallowed them to establish for at least 1 weekbefore experimenting (17). The growth rates[analysis of variance (ANOVA), F3,118 �1.837, P � 0.144] and morphology of the threetransformed plant lines did not differ from thoseof wild-type (WT) plants under the noncompet-itive conditions of the experiment (26). Theplants, however, did differ in their productionof VOCs (17).

The release of herbivore-induced terpe-noid compounds, in particular the sesquiter-pene, cis-�-bergamotene, is a demonstratedsentinel for other herbivore-induced changesin oxylipin-dependent secondary metaboliteproduction of N. attenuata plants (10, 17, 18).Undamaged plants of all four genotypes re-leased equally small amounts of cis-�-bergamotene (ANOVA, F3,11 � 0.876, P �0.483). WT plants and HPL-deficient plantsreleased significantly increased amounts ofcis-�-bergamotene and a suite of other terpe-noid compounds after M. sexta damage (tableS1). In contrast, the cis-�-bergamotene emis-sion from herbivore-damaged as-lox and as-aos (ANOVA, F3,10 � 44.56, P � 0.0001;Bonferroni post hoc P � 0.05) plants re-mained low (Fig. 1A).

In addition to the open-flow VOC trap-ping design, we used a portable gas chro-

matograph (z-Nose) (27) to characterize thewound-induced emissions of the GLV, cis-3-hexenal, from WT and transformed plants(17). As-hpl plants released significantly lesscis-3-hexenal than did WT, as-aos, and as-loxlines immediately after damaging the leaftissues (ANOVA, F3,19 � 33.07, P � 0.0001;Bonferroni post hoc P � 0.05) (Fig. 1B). Thisdemonstrates that the production of GLVs,such as cis-3-hexenal, requires the activity ofHPL but not LOX3, which specifically sup-plies fatty acid hydroperoxides to the octade-canoid pathway but not to the HPL reaction(17, 18) (fig. S1). Interestingly, AOS-deficient plants had significantly higher cis-3-hexenal emission levels after the mechani-cal damage than did all other plant lines(Bonferroni post hoc P � 0.05) (Fig. 1B),which suggests either a rechanneling of AOSsubstrates into the HPL cascade (20) and/oran octadecanoid-derived negative regulationof the HPL cascade.

The altered production of secondary metab-olites, such as terpenoids, in as-lox plants wascorrelated with reduced resistance to attack bythe specialist herbivore M. sexta. Freshlyhatched M. sexta caterpillars, which fed onrosette-stage, field-grown WT plants and simi-lar-sized plants of the three different trans-formed lines (17), gained weight fastest onLOX3-deficient plants (ANOVA, F3,23 � 5.83,P � 0.0041). After 9 days of development onas-lox plants, they were 4.4-fold heavier thancaterpillars on WT plants and 7-fold and 2.5-fold heavier than those on HPL- and AOS-deficient plants (Bonferroni post hoc tests, P �0.05), respectively (Fig. 1C). In laboratory ex-periments, herbivore resistance can be restoredby treating as-lox plants with methyl jasmonate(20). The relative resistance of AOS-deficientplants to herbivores and the susceptibility ofLOX3-deficient plants confirm laboratory re-sults and correlated with the wound-inducedproduction of nicotine. Nicotine production wasreduced only in as-lox plants, which was attrib-uted to a leaky phenotype in the as-aos plants(fig. S1B) (17, 20). Similarly, as-hpl plantsretain their resistance to hornworm damage de-spite the evidence that HPL-derived oxylipinsmay signal plant-defense activation (22, 23). Inlaboratory studies, hornworm consumption andgrowth were slower on HPL-deficient plantsthan on WT plants but could be restored to WTlevels by the addition of GLVs to as-hpl plants,which suggested that GLVs stimulate feedingby M. sexta (20).

The initial field experiments showed that thethree transformed plant lines (as-lox, as-aos,as-hpl) have similar characteristics in the fieldand in the laboratory (18, 20) and differ in theirresponsiveness to M. sexta attack. However, theherbivore community of N. attenuata is diverseand includes piercing-sucking, leaf-mining, andleaf-chewing herbivores, such as M. sexta.Moreover, the composition of the herbivore

community can be extremely variable and isinfluenced by random effects (5, 9, 10). There-fore, the exposure of genetically manipulat-ed plants to their natural herbivore commu-nity realistically evaluates the role playedby oxylipin-mediated herbivore-inducedresponses in structuring the plants’ herbi-vore community.

Fig. 1. Volatile organic compound emissionand herbivore susceptibility of N. attenuatawild-type (WT) plants and plants trans-formed to silence LOX3 (as-lox), HPL (as-hpl), and AOS (as-aos) activity. (A) Mean(�SEM) cis-�-bergamotene emission (nghour�1) from undamaged (CON) andManduca sexta hornworm-damaged (HD)plants of the four genotypes. (B) Mean(�SEM) emission of cis-3-hexenal (ng g�1

fresh mass) measured with a portable gaschromatograph (z-Nose) in the headspace ofexcised leaves that had been mechanicallywounded. (C) Mean (�SEM) caterpillar mass(g) after 9 days of development on WT plantscompared to the genetically transformedlines. Different letters designate significantlydifferent means as informed by a Bonferronipost hoc (P � 0.05) test of an ANOVA.

R E P O R T S


We transplanted 80 young rosette-stageplants in groups of four into a field plot along alinear transect (17). Each group comprised oneWT plant and one plant of each of the trans-formed lines: as-lox, as-hpl, and as-aos. Exten-sive prior laboratory analysis of plants trans-formed with empty-vector constructs revealedno differences in any measured herbivore resis-tance trait or growth and reproduction whencompared to WT plants (16). During the 2002–2003 field seasons, we observed damage fromherbivores of all three feeding guilds but foundvery few plants that were attacked by the mostdamaging herbivores of previous field sea-sons: the leaf-chewing sphingid hornwormsManduca quinquemaculata and M. sexta and thepiercing-sucking mirid bug Tupiocoris notatus(5, 10). Thus the observed damage, though rel-atively low, allowed the distribution of herbivoryon the silenced lines to be clearly evaluated.

If plants within a population show induceddefense responses to herbivory and if herbi-vores can freely choose among plants, her-bivory should be equally distributed. Plantslacking the ability to respond to herbivore at-tack are more susceptible and should be pre-ferred to responsive plants (7, 8). Consistentwith this prediction, we found dramaticallymore herbivore damage on as-lox plants(ANOVA, F3,120 � 24.19, P � 0.0001; Bon-ferroni post hoc P � 0.05) than on WT, as-hpl,or as-aos plants (Fig. 2A, fig. S2A), whichcorresponds to the increased performance of M.sexta caterpillars on as-lox plants. Similarly,herbivores attacked a significantly greater pro-portion of as-lox plants compared to all otherlines in the experimental population (Fig. 2A).Although herbivory was equally distributedamong WT, as-aos, and as-hpl plants (Pearsonchi-square test �2 � 2.518, P � 0.05), it wassignificantly higher on plants with strongly at-tenuated induced responses (as-lox) (Pearsonchi-square test �2 � 49.207, P � 0.0001),which suggests that a plant’s ability to elicitdefenses influences the distribution of her-bivory within a plant population.

Moreover, a more detailed analysis of theherbivore community revealed that theherbivore-induced plant responses can alterthe host breadth of generalist herbivores.Compared to previous study years, we foundtwo new herbivores on the experimentalplants: the leafhopper Empoasca sp. (28)(Fig. 3, fig. S2B) and the western cucumberbeetle Diabrotica undecimpunctata tenellaLe Conte (fig. S2C). In fact, most of theobserved damage on LOX3-deficient plantsresulted from just one of these, Empoasca

sp., which is an unusual herbivore on N.attenuata plants. In 12 quantitative formalsurveys of 600 plants conducted over fourconsecutive years, we found a total of onlyfour Empoasca sp. specimens on N. attenuataplants growing in native populations (tableS2). Neither in these surveys, nor in informalobservations of more than 40,000 plants overthe past 15 years of field work has there beenevidence of Empoasca feeding damage. Em-poasca sp. is a highly mobile, opportunisticherbivore that attacks Datura wrightii, Sola-num americanum, and Mirabilis multiflora,all of which were abundant in the study area.In our field trials with different N. attenuatalines, the as-lox plants received significantlymore Empoasca damage than did the othertransformed lines (as-aos, as-hpl) or the WT(ANOVA, F3,121 � 30.25, P � 0.0001, Bon-ferroni post hoc test P � 0.05) (Fig. 2B).More than 68% of the as-lox plants in theexperimental population were damaged byEmpoasca, whereas the proportion of dam-aged plants in the other lines did not exceed29% (Fig. 2B). More importantly, Empoascafemales oviposited on as-lox plants, and theseplants were susceptible to attack from theemerging leafhopper offspring (Fig. 3B). Em-poasca nymphs were found on 81% of theattacked as-lox plants but not at all on WT,HPL-, and AOS-deficient lines. Moreover,when given the choice (17), adult leafhoppersclearly prefer as-lox plants over WT plants(paired Student’s t test, t � �4.919, P �0.0012) (Fig. 4). Interestingly, we found noEmpoasca damage on as-hpl plants, suggest-ing that leafhopper feeding behavior may alsobe influenced by GLVs, as has been demon-strated with M. sexta larvae (20).

The reduced resistance of LOX3-deficientplants was correlated with an alteredoctadecanoid-dependent gene expression inresponse to Empoasca attack. With a cDNAmicroarray analysis, we compared transcriptaccumulation in response to Empoasca feed-ing in WT and as-lox plants to identify

Fig. 2. Herbivore damage on field-grown N.attenuata wild-type (WT) plants and plantstransformed to silence LOX3 (as-lox), HPL (as-hpl), and AOS (as-aos) activity. (A) Mean(�SEM) percentage of leaf area damaged byherbivores (bar graph) and proportion of plantsin the population that were attacked by herbi-vores (white in the pie chart) or remainedundamaged (gray in the pie chart). (B) Mean(�SEM) percentage of leaf area damaged byEmpoasca sp. leafhoppers (bar graph) and pro-portion of plants in the population that wereattacked by leafhoppers (white in the pie chart)or remained undamaged (gray in the pie chart).Different letters designate significantly differ-ent means as informed by a Bonferroni posthoc (P � 0.05) test of an ANOVA.

Fig. 3. Empoasca sp. leafhoppers on as-lox N.attenuata plants. (A) Adult Empoasca sp. (B)Empoasca sp. nymphs were exclusively foundon as-lox plants. (C) Visual damage by Em-poasca sp. on as-lox N. attenuata plants.

Fig. 4. Mean (�SEM) leaf tissue damage onwild-type (WT) and as-lox plants after 3 daysof attack by Empoasca sp. leafhoppers or D.undecimpunctata leaf beetles, respectively. TenEmpoasca and two D. undecimpunctata, re-spectively, were allowed to choose between aWT and an as-lox plant that had been pottedtogether and covered with insect mesh. Bothexperiments were replicated 10 times.

R E P O R T S


octadecanoid-dependent transcriptional re-sponses (fig. S3) (17). Of the 240 N. attenu-ata genes spotted on the microarray, 54 genesshowed significantly increased LOX3-dependent transcript accumulation after Em-poasca damage and 50 showed decreasedaccumulation (table S3). A series of defense-related genes, such as trypsin proteinase in-hibitor (TPI) and threonine deaminase, hadincreased expression levels, whereas others,such as xyloglucan endotransglucosylase/hydrolase or GAL83 and many of thephotosynthesis-related genes (e.g., small sub-unit of ribulose bisphosphate carboxylase),were down-regulated in response to Em-poasca attack and showed little if any regu-lation in as-lox plants (table S3). Theseresults suggest that a complex LOX3-dependent regulation of primary and second-ary metabolism in N. attenuata mediates re-sistance to the piercing-sucking Empoascaleafhoppers. The fact that octadecanoid sig-naling plays a role in plant defense againstboth piercing-sucking and leaf-chewing in-sects suggests a common response to attackfrom members of these two feeding guilds.Moreover, LOX3-dependent octadecanoidsmay play a direct role in host-plant selectionby enabling herbivores to differentiate be-tween plants with and without intact JA sig-naling, as shown in experiments with the cornearworm, Helicoverpa zea. This generalistherbivore uses induced JA and salicylate toactivate four of its cytochrome P450 genesthat are associated with detoxification eitherbefore or concomitantly with the biosynthesisof allelochemicals (29).

In addition to M. sexta, which was af-fected by LOX3-mediated plant resistancetraits, we found a novel leaf-chewing her-bivore on N. attenuata, the leaf beetle Di-abrotica undecimpunctata tenella Le Conte(fig. S2C). It often feeds on D. wrightii andCucurbita foetidissima flowers in the studyarea and was observed on N. attenuataplants exclusively in this study and only onas-lox plants. To test whether or not thisclear preference in the field is caused by thedecreased expression of LOX3, we allowedthe beetles to choose between WT and as-lox plants. The choice experiment revealeda clear preference for as-lox plants com-pared to WT plants (paired Student’s t test,t � �4.050, P � 0.003) (Fig. 4).

Our results show that the LOX3-mediatedinducibility of plants is crucial for the ovipo-sition decision and for the opportunistic hostselection behavior of Empoasca sp. and D.undecimpunctata, and thereby defines hostbreadth. Therefore, host selection seemsdetermined not only by the plant’s constitu-tively expressed chemical phenotype and ex-ternal mortality factors (predation pressure,abiotic stress) (30) but also by the plant’sability to induce responses to herbivory. The

additional finding that induced responses toherbivory influence the distribution of her-bivory within a plant community points to thevalue of genetically silenced plants in ecolog-ical research. An understanding of the eco-logical interactions that occur in nature isessential for sustainable agriculture.

References and Notes1. S. S. Duffey, M. J. Stout, Arch. Insect Biochem. Physiol.

32, 3 (1996).2. C. M. De Moraes, W. J. Lewis, P. W. Pare, H. T. Alborn,

J. H. Tumlinson, Nature 393, 570 (1998).3. M. Dicke, J. J. A. van Loon, Entomol. Exp. Appl. 97,

237 (2000).4. J. S. Thaler, M. J. Stout, R. Karban, S. S. Duffey, Ecol.

Entomol. 26, 312 (2001).5. A. Kessler, I. T. Baldwin, Science 291, 2141 (2001).6. T. C. J. Turlings, J. H. Tumlinson, W. J. Lewis, Science

250, 1251 (1990).7. R. Karban, I. T. Baldwin, Induced Responses to Her-

bivory (Univ. of Chicago Press, Chicago, IL, 1997).8. A. A. Agrawal, Science 294, 321 (2001).9. C. A. Preston, I. T. Baldwin, Ecology 80, 481 (1999).

10. A. Kessler, I. T. Baldwin, Plant J. 38, 639 (2004).11. R. Halitschke, U. Schittko, G. Pohnert, W. Boland, I. T.

Baldwin, Plant Physiol. 125, 711 (2001).12. E. Blee, Trends Plant Sci. 7, 315 (2002).13. R. A. Creelman, J. E. Mullet, Annu. Rev. Plant Physiol.

Plant Mol. Biol. 48, 355 (1997).14. J. Bergelson, C. B. Purrington, C. J. Palm, J. C. Lopez-

Gutierrez, Proc. R. Soc. London B Biol. Sci. 263, 1659(1996).

15. J. Royo et al., Proc. Natl. Acad. Sci. U.S.A. 96, 1146(1999).

16. J. A. Zavala, A. G. Patankar, K. Gase, I. T. Baldwin,Proc. Natl. Acad. Sci. U.S.A. 101, 1607 (2004).

17. Materials and methods are available as supportingmaterial on Science Online.

18. R. Halitschke, I. T. Baldwin, Plant J. 36, 794 (2003).19. L. Li et al., Plant Cell 16, 126 (2004).20. R. Halitschke, M. Keinanen, J. Ziegler, I. T. Baldwin,

Plant J., in press.

21. D. F. Hildebrand, G. C. Brown, D. M. Jackson, T. R.Hamilton-Kemp, J. Chem. Ecol. 19, 1875 (1993).

22. N. J. Bate, S. J. Rothstein, Plant J. 16, 561 (1998).23. G. Arimura, R. Ozawa, J. Horiuchi, T. Nishioka, J.

Takabayashi, Biochem. Syst. Ecol. 29, 1049 (2001).24. J. H. Park et al., Plant J. 31, 1 (2002).25. A. Stinzi, J. Browse, Proc. Natl. Acad. Sci. U.S.A. 97,

10625 (2000).26. As plants began to elongate and produce flowers,

they were examined daily; all flowers were removedbefore opening and anthesis to meet the perfor-mance standards determined in the Code of FederalRegulations [7CFR340.3(c)] for the Introduction ofOrganisms Altered or Produced through Genetic En-gineering. Consequently, direct fitness measureswere unattainable in these experiments.

27. M. Kunert, A. Biedermann, T. Koch, W. Boland, J. Sep.Sci. 25, 677 (2002).

28. Probably belonging to the species E. fabae (Harris,1841), which is widely known as a pest on alfalfa inthe United States.

29. X. C. Li, M. A. Schuler, M. R. Berenbaum, Nature 419,712 (2002).

30. E. Bernays, M. Graham, Ecology 69, 886 (1988).31. Supported by the Max-Planck-Gesellschaft and

Deutsche Forschungsgemeinschaft (BA-2132/1-1).We thank E. Wheeler, E. Pichersky, D. Heckel, J.Gershenzon, M. Heil, and H. Vogel for helpful com-ments; R. Baumann, P. Freytag, and H. Nickel forassistance with species determination; and BrighamYoung University for use of Lytle Preserve as a fieldstation; L. Rausing for helping us promote the discus-sion of the scientific value of transformed plants; andJ. White and the Animal and Plant Health InspectionService personnel for facilitating their safe use innature.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/1096931/DC1Materials and MethodsFigs. S1 to S3Tables S1 to S3References

18 February 2004; accepted 7 June 2004Published online 1 July 2004;10.1126/science.1096931Include this information when citing this paper.

Osedax: Bone-Eating MarineWorms with Dwarf MalesG. W. Rouse,1,2* S. K. Goffredi,3* R. C. Vrijenhoek3†

Wedescribe a new genus,Osedax, and two new species of annelids with femalesthat consume the bones of dead whales via ramifying roots. Molecular andmorphological evidence revealed that Osedax belongs to the Siboglinidae,which includes pogonophoran and vestimentiferan worms from deep-sea vents,seeps, and anoxic basins. Osedax has skewed sex ratios with numerous dwarf(paedomorphic) males that live in the tubes of females. DNA sequences revealthat the two Osedax species diverged about 42 million years ago and currentlymaintain large populations ranging from 105 to 106 adult females.

Deep-sea exploration continues to reveal bi-ological novelties (1) such as whale fall com-munities (2). Here, we describe remarkablepolychaete annelids, Osedax gen. nov. (nov.),discovered in January 2002 on the bones of agray whale carcass at 2891 m depth inMonterey Bay, California (3). Their conspic-uous red plumes extended from most exposedportions of the whale bones (Figs. 1A and2A). Colonies of these worms comprised twospecies, Osedax rubiplumus sp. nov. and O.

frankpressi sp. nov., that we describe alongwith the new genus. Nucleotide sequenceanalysis revealed that the two Osedax speciesdiffered by 17.28 � 0.21% (x � SD) for mito-chondrial COI, by 7.63 � 0.46% for mito-chondrial 16S rRNA, and by 4.09 � 0.04%for nuclear 18S rRNA (4). On the basis of amolecular clock calibrated for COI in deep-sea annelids (5), O. rubiplumus and O. frank-pressi diverged about 42 million years ago(Ma) (4 ), in the late Eocene, when their

R E P O R T S


ancestor may have exploited the bones ofarcheocete whales such as Basilosaurus(6 ). Phylogenetic analysis (Fig. 3) placedOsedax in the family Siboglinidae (7, 8),which includes frenulate and vestimentif-eran tubeworms that also lack digestivesystems (9, 10).

Unlike other siboglinids, female Osedaxlack a discrete trophosome, the organ housingsymbiotic bacteria in vestimentiferans andpogonophorans. Instead, Osedax possess a bul-bous posterior ovisac covered by a sheath ofgreen-colored tissue that branches into a vascu-larized “root” system and invades the bonemarrow (Figs. 1, C and H, and 2, C, D, and F).This branching root system is histologicallydistinct and not homologous with the singularchitinous root tube found in some Lamellibra-chia vestimentiferans (11). Microscopic andmolecular analyses (12) of this sheath revealedbacteriocytes (Fig. 1J) containing large rod-shaped bacteria of the microbial order Oceano-spirillales, known for heterotrophic degradationof complex organic compounds. Analyses ofstable isotopes and fatty acids (12) revealed thatthe endosymbionts are responsible for the nu-trition of this worm. This heterotrophic symbi-osis differs markedly from the chemolitho-autotrophic symbioses found in other deep-seaannelids and mollusks that rely on sulfide- ormethane-oxidizing bacterial endosymbionts(13). This finding of an endosymbiosis involv-ing heterotrophic degradation in Osedax sug-gests that the evolutionary history of bacterialsymbioses among the Siboglinidae is more var-ied than previously suspected (14). Reliance onthe bones of marine mammals, hydrocarbondegradation, and the unusual morphology of thesymbiont-bearing ovisac and root system ofthese worms make this particular symbioticassociation unique in the animal kingdom.

All Osedax visible to the eye were femalesthat ranged from 0.2 to 0.5 mm in trunk width,suggesting ongoing recruitment. Females assmall as 0.3 mm wide produced eggs (Figs. 1,D and F, and 2F). The tubes of individualfemales contain numerous microscopic males(Figs. 1, K to O, and 2G) that were filled withdeveloping sperm (Fig. 1N) and often containedyolk droplets (Fig. 2I). These paedomorphicmales retain morphological traits typical of si-boglinid trochophore larvae (15), including aciliary band that appears to be a putative pro-totroch (Figs. 1L and 2I) and opisthosomalchaetae (Figs. 1O and 2H). The tubes of largerfemales contained up to 111 males each, with a

male-to-female sex ratio of 17:1. Females ei-ther accumulate males over time or attract morewhen larger, because the number of males iscorrelated (r � 0.899, P � 0.01) with femalesize in O. rubiplumus (16). We hypothesize thatsex may be environmentally determined inOsedax, with larvae settling on exposed bonesmaturing as females and those landing on fe-males becoming males. Environmental sex de-termination is known in the echiuran Bonellia(17), now regarded as a polychaete (18).

Amounts of mitochondrial COI diversity(�) suggest that the effective female popula-tion sizes [Ne(f)] range from 5 � 105 in O.rubiplumus to 9 � 105 in O. frankpressi.

These numbers are of the same magnitude asestimates of Ne(f) (range of 105 to 106 fe-males) inferred for deep-sea annelids (4) andconsistent with estimates of Ne(f) from otherinvertebrates (19). The large female popula-tion sizes estimated for these Osedax speciessuggest they are common on whale falls.These numbers also suggest that the frequen-cy of whale falls has historically been great,which is consistent with estimates of largewhale populations before modern whaling (2,20). Their abundance suggests that theseworms might play a substantial role in thecycling of large organic inputs into the sur-rounding deep-sea communities.

1South Australian Museum, North Terrace, AdelaideSA 5000, Australia. 2Earth and Environmental Scienc-es, University of Adelaide, Adelaide SA 5005, Austra-lia. 3Monterey Bay Aquarium Research Institute(MBARI), 7700 Sandholdt Road, Moss Landing, CA95039, USA.

*These authors contributed equally to the work.†To whom correspondence should be addressed. E-mail: [email protected]

Fig. 1. O. rubiplumus. (A) Bones in situ with emergent worms (arrows). (B) Whale rib with femaleworms. (C) Female, tube removed, dissected from bone. (D) Anterior of oviduct with eggs. (E)Dorsal crown-trunk junction. (F) Lateral trunk base with oviduct emerging from ovisac (arrow). (G)Transverse section of anterior trunk. Male sits on oviduct (arrow). (H) Lateral ovisac and roots ofholotype. (I) Dwarf males (arrows) in tube with female. (J) Transmission electron micrograph (TEM)of root tissue with bacteriocytes. (K) Dwarf male with sperm duct (arrow) and spermatids. (L)Scanning electron micrograph (SEM) of male anterior. (M) Longitudinal section (LS) through maleanterior. (N) Longitudinal TEM section through posterior of sperm head. (O) SEM of posterior hooks(arrows) on male. (P) LS through hook. (Q) SEM of hook with subrostral teeth (arrow). b, bone; babacteria; c, collar; cm, circular muscle; dbv, dorsal blood vessel; od, oviduct; lm longitudinal muscle;m, male; op, ovisac projection; os, ovisac; p, palps; pl, plaques; pr, prototroch; r, roots; sd, spermduct; st, spermatids; t, trunk; tu, tube; and vbv, ventral blood vessel.

R E P O R T S


Systematic description. Annelida, Lamarck1809; Canalipalpata, Rouse and Fauchald 1997;Siboglinidae Caullery, 1914. Osedax gen. nov.Diagnosis. Polychaete worms with females hav-ing a discrete red crown, contractile trunk, bul-bous ovisac, and branching roots. Crown andtrunk within transparent tube emergent fromwhale bone (Figs. 1, A and B, and 2, A to C).Crown composed of cylindrical oviduct (Figs.1D and 2E) plus four pinnule-bearing palps(Figs. 1C and 2, D and E). No mouth or obviousgut. Cylindrical trunk composed mostly of lon-gitudinal muscles and glands. Dorsal heart lies atanterior region of trunk; major dorsal and ventralblood vessels present (Fig. 1G). Oviduct paral-lels trunk surface into posterior ovisac (Figs. 1, Cand F to G, and 2D) filled with numerous whiteoocytes. Ovisac enclosed by green sheath (Fig.2F) composed of epidermis and bacteriocytescontaining bacteria (Fig. 1J). Ovisac sheath con-tinuous with variably branching posterior roots(Figs. 1, C and H, and 2D). Vascularized rootsand ovisac (Fig. 2F). No chaetae or segmentationapparent in females. Paedomorphic males clusteraround oviduct in female tubes (Figs. 1, I and K,and 2G). Males with anterior prototroch (Figs.1L and 2I) and posterior hooked chaetae (Figs.1O and 2H) arranged in two rows of four pairs(Fig. 1O). Hooks, lacking rostrum, comprise ca-pitium with curved teeth over subrostral pro-cess (Figs. 1, P and Q, and 2H). Internally,males contain spermatids and sperm in anteri-or duct (Fig. 1, K and M, and 2I). Etymology.From Latin os, bone, and edax, devouring;gender masculine. Type species. Osedax

rubiplumus sp. nov. by present designation.Osedax rubiplumus sp. nov. Type mate-

rial. (21) Monterey Bay, California, Tiburondive T486, 36°36.8’N, 122°26.0’W, 2891 m,8 February 2002: holotype, mature adult fe-male (CASIZ 170238); allotypes, 38 malesfrom tube of holotype (CASIZ 170240); para-types, 10 females and numerous males(CASIZ 170239), 10 females and numerousmales (LACM-AHF POLY 02146), and 31females and numerous males (SAM E3376).Diagnosis. Holotype, emergent body in cy-lindrical tube, walls 1 mm thick. Contractedcrown plumes 2.1 cm long. Oviduct, filledwith ellipsoid eggs (mean diameters 151 �mby 121 �m, n � 30). Oviduct extends be-tween palps, 1.8 cm from trunk. Palps red inliving worms, pinnules on outer margins.Collar with brown plaques dorsally (Fig. 1E).Trunk 3.8 cm long, 2 mm wide at collar.Ovisac, 8 mm by 4 mm by 0.3 mm, withpaired anterolateral projections (Fig. 1C) andfour discrete roots posteriorly. Roots withspherical lobes (Fig. 1H). Allotypes 0.4- to1.1-mm-long males. Putative prototroch in-complete (Fig. 1L). Posteriorly, 16 hookswith capitium teeth emergent; handles 18 to23 �m (Fig. 1P). Capitium with six to eightteeth; two smaller subrostrumal teeth (Fig.1Q). Etymology. From Latin rubi, red, andpluma, feather.

Osedax frankpressi sp. nov. Type material.(21) Monterey Bay, California, Tiburon diveT610, 36°36.8’N, 122°26.0’W, 2891 m, 7 Au-gust 2002: holotype, mature adult female

(CASIZ 170235); allotypes, 80 males from tubeof holotype (CASIZ 170237); paratypes, threefemales (CASIZ 170236), three females(LACM-AHF POLY 02147), three females(SAM E3377). Diagnosis. Holotype, emergentbody in gelatinous hemispherical tube, 7-mmdiameter. Contracted crown plumes 0.95 cmlong. Oviduct filled with ellipsoid eggs(mean diameters 146 �m by 117 �m; n �15), Oviduct convoluted upon contraction,extending between palps, 3 mm from trunk(Fig. 2, D and E). Palps red with two longi-tudinal white stripes in living worms (Fig.2A); pinnules on inner margins. Trunk 4.5mm long, 0.9 mm wide, and marked bywhite thickened tissue at anterior (Fig. 2E).Green, bacteriocyte-filled sheath formstrunk-ovisac junction 1.2 mm long, 1 mmwide (Fig. 2D). Lobulate ovisac, 6.5 mm by5 mm by 3 mm (Fig. 2F). Ovisac and rootsinflated with clear fluid in situ in bone (Fig.2C); fluid lost on extraction. Allotypes,0.15- to 0.25-mm-long males (Fig. 2, G andI). Chaetae with hooks and handles 15 to 21�m (Fig. 2H). Capitium with five teeth; nosubrostral teeth (Fig. 2H). Etymology. Inhonor of Dr. Frank Press, former U.S. pres-idential science advisor, president of theU.S. National Academy of Sciences, andchair of the MBARI Board of Directors, forhis distinguished service to science.

Remarks. Females of the two new speciesare easily distinguished by the lengths of theirtubes and palp coloration. O. rubiplumus tubeshave a uniform diameter and maintain shapewhen removed from the water, whereas O.frankpressi tubes are gelatinous and collapse inair. The white-striped O. frankpressi palps con-trast with the uniform red palps of O. rubiplu-mus. Females of the two species differ in the

Fig. 2. O. frankpressi. (A) Whale rib in situ with emergent worms. (B) Retracted worms. (C) Femalepartly dissected from bone; note fluid-filled ovisac. (D) Female dissected from bone. (E) Crown-trunk junction. (F) Ovisac with green sheath cut to reveal ovary and blood vessels. (G) Dwarf malesin female tube. (H) Male hooks. (I) Male showing prototroch, developing sperm, and yolk. bv, bloodvessel; y, yolk; otherwise as for Fig. 1.

Fig. 3. Osedax as member of Siboglinidae. Bayes-ian analyses of 2088 molecular characters fromcombined 16S and 18S rDNA (22). Posterior prob-ability values of 100% indicated by asterisks.

R E P O R T S


trunk-ovisac junction, ovisac shape, and rami-fication of roots. Males of O. frankpressi areless than one-third the size of O. rubiplumusmales. Male chaetae of O. rubiplumus havesubrostral teeth and more capitium teeth.

References and Notes1. C. L. Van Dover, C. R. German, K. G. Speer, L. M.

Parson, R. C. Vrijenhoek, Science 295, 1253 (2002).2. C. R. Smith, A. R. Baco, Oceanogr. Mar. Biol. Annu.

Rev. 41, 311 (2003).3. S. K. Goffredi, C. K. Paull, Fulton-Bennett, L. A. Hur-

tado, R. C. Vrijenhoek, Deep Sea Res. I, in press.4. Details on materials and methods and GenBank ac-

cession numbers are available as supporting materialon Science Online.

5. P. Chevaldonne, D. Jollivet, D. Desbruyeres, R. A. Lutz, R. C.Vrijenhoek, Cah. Biol. Mar. 43, 367 (2002).

6. P. D. Gingerich, B. H. Smith, E. L. Simons, Science 249,154 (1990).

7. K. M. Halanych, R. A. Feldman, R. C. Vrijenhoek, Biol.Bull. 201, 65 (2001).

8. G. W. Rouse, Zool. J. Linn. Soc. 132, 55 (2001).9. E. C. Southward, in Microscopic Anatomy of Inverte-

brates, vol. 12 of Onychophora, Chilopoda and LesserProtostomata, F. W. Harrison, M. E. Rice, Eds. (Wiley-Liss, New York, 1993), pp. 327–369.

10. M. L. Jones, Science 213, 333 (1981).11. D. Julian, F. Gaill, E. Wood, A. Arp, C. Fisher, J. Exp.

Biol. 202, 2245 (1999).12. S. K. Goffredi et al., in preparation.13. C. L. Van Dover, The Ecology of Deep-Sea Hydrother-

mal Vents (Princeton Univ. Press, Princeton, NJ,2000), p. 411.

14. A. Schulze, K. M. Halanych, Hydrobiologia 496, 199 (2003).15. E. C. Southward, Hydrobiologia 402, V185 (1999).16. Fifty-four female Osedax rubiplumus from a piece of

rib obtained on Tiburon dive 486 were measured andassessed for spawning, and their tubes were checkedfor males. Details are available at (4).

17. J. Pilger, in Settlement and Metamorphosis of MarineInvertebrate Larvae, F.-S. Chia, M. E. Rice, Eds.(Elsevier, New York, 1978), pp. 103–112.

18. D. McHugh, Proc. Natl. Acad. Sci. U.S.A. 94, 8006 (1997).19. M. Lynch, J. S. Conery, Science 302, 1401 (2003).20. J. Roman, S. R. Palumbi, Science 301, 508 (2003).21. Repositories of type specimens are as follows: CASIZ,

California Academy of Sciences; LACM, Los AngelesCounty Museum; and SAM, South Australia Museum.

22. Phylogenetic analysis has its basis in 18S and 16S rDNAsequences. The amphinomid Hipponoe was chosen to rootthe tree. Additional polychaete sequences were obtainedfrom GenBank. Ambiguously aligned sections of 18SrDNA were excluded from the analysis. PAUP*4.0b10(D. Swofford, Florida State University) was used formaximum parsimony analysis with characters equallyweighted and gaps treated as missing data. Bootstrapvalues were estimated with the use of 1000 heuristicsearches. Bayesian analysis used MrBayes v3.0B4 (J.Huelsenbeck, University of Rochester, NY) with parti-tions for stems and loops using RNA secondary struc-ture prediction via GeneBee (L. Brodsky, Moscow StateUniversity, Russia). Likelihood models were chosen withthe use of ModelTest 3.06 (D. Posada, Universidad deVigo, Spain). Further details are available at (4).

23. Funding provided by the David and Lucile PackardFoundation, NSF (OCE9910799 and OCE0241613),and the South Australian Museum. Thanks to Tiburonpilots and Western Flyer crew for obtaining samples;J. Jones and R. Young for help with Bayesian analysis;R. Newbold and F. Pleijel for Latin advice; H. Schoppe,K. Rogers, and Adelaide Microscopy for microscopysupport, and V. Orphan, L. Jahnke, T. Embaye, and K.Turk (NASA Ames Research Center) for nutritionalcharacterization of the worms.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/668/DC1Materials and MethodsFig. S1Tables S1 to S3


The Complete Genome Sequenceof Propionibacterium Acnes,a Commensal of Human SkinHolger Bruggemann,1*† Anke Henne,1 Frank Hoster,1

Heiko Liesegang,1 Arnim Wiezer,1 Axel Strittmatter,1

Sandra Hujer,2 Peter Durre,2 Gerhard Gottschalk1†

Propionibacterium acnes is a major inhabitant of adult human skin, where itresides within sebaceous follicles, usually as a harmless commensal althoughit has been implicated in acne vulgaris formation. The entire genome sequenceof this Gram-positive bacterium encodes 2333 putative genes and revealednumerous gene products involved in degrading host molecules, including siali-dases, neuraminidases, endoglycoceramidases, lipases, and pore-forming fac-tors. Surface-associated and other immunogenic factors have been identified,which might be involved in triggering acne inflammation and other P. acnes–associated diseases.

The details of the involvement of Propi-onibacterium acnes in acne—the most com-mon skin disease, affecting up to 80% of alladolescents in the United States—are stillobscure. Several mechanisms have been pro-posed to account for its role in the disease(1–5). First, damage to host tissues and cellsmight be accomplished by bacterial enzymeswith degradative properties, such as lipases(2). Second, immunogenic factors of P. acnessuch as surface determinants or heat shockproteins (HSPs) might trigger inflammation(4–6). Other diseases are also associated withP. acnes, including corneal ulcers; endocar-ditis; sarcoidosis; cholesterol gallstones; al-lergic alveolitis; pulmonary angitis; and sy-novitis, acne, pustulosis, hyperostosis, andosteitis (SAPHO) syndrome (7, 8). Its ge-nome sequence may provide a basis for find-ing alternative targets in therapy for acne andother P. acnes–associated diseases.

The genome of P. acnes strain KPA171202(no. DSM 16379) consists of a single circularchromosome of 2,560,265 base pairs (9, 10)(supporting online text and fig. S1). We predict-ed and annotated 2333 putative genes. Thesequenced strain exhibited 100% identity onthe 16S ribosomal RNA level to several clin-ical P. acnes isolates, as well as to the well-studied laboratory strain P-37. The main fea-tures of the genome sequence and comparativeanalyses are described in the supporting onlinematerial (supporting online text and fig. S2).

The genome sequence offers insights into thetraits that favor P. acnes as a ubiquitous com-mensal on human skin. Metabolic reconstructionreveals a capacity to cope with changing oxygentensions, which confirms observations thatstrains of P. acnes can grow under microaerobicas well as anaerobic conditions. The genomesequence encodes all key components of oxida-tive phosphorylation that employs two terminaloxidases, a cytochrome aa3 oxidase (PPA701/702) and a cytochrome d oxidase (PPA173-176),and a F0F1-type adenosine triphosphate synthase(PPA1238-1245). All genes of the Embden-Meyerhof pathway, the pentose phosphate path-way, and the tricarboxylic acid cycle are present.Under anaerobic conditions, strain KPA171202can grow on several substrates such as glucose,ribose, fructose, mannitol, trehalose, mannose,N-acetylglucosamine, erythritol, and glycerol(11). In addition, several amino acid degradingpathways, similar to those of fermentative organ-isms, are present. Fermentative products areshort-chain fatty acids, in particular propionicacid (11), whose production from pyruvateand methylmalonyl–coenzyme (CoA) is initial-ized by the methylmalonyl-CoA carboxyltrans-ferase (PPA2005-2008). In addition to fermen-tative energy conservation, P. acnes possessessystems involved in anaerobic respiration suchas nitrate reductase (PPA507-511), dimethyl sul-foxide reductase (PPA515-517), sn-glycerol-3-phosphate dehydrogenase (PPA2248-2250), andfumarate reductase (PPA950-952 and PPA1437-1439). Factors that might be involved in thelife-style switch related to oxygen availability, aswell as further aspects of the biology and bio-chemistry of P. acnes, are presented in the sup-porting online text.

Numerous genes have been found that candegrade and use host-derived substances (Table1 and fig. S3). It has been proposed that freefatty acids, produced by P. acnes lipase activityon sebum, assist bacterial adherence and colo-nization of the sebaceous follicle (2, 12). In

1Gottingen Genomics Laboratory, Institute of Micro-biology and Genetics, Georg-August-University Got-tingen, Grisebachstra�e 8, 37077 Gottingen, Germa-ny. 2Department of Microbiology and Biotechnology,University of Ulm, 89069 Ulm, Germany.

*Present address: Institut Pasteur, Laboratory ofGenomics of Microbial Pathogens, 28 Rue du Dr. Roux,75724 Paris, France.†To whom correspondence should be addressed. E-mail: [email protected] (H.B.) and [email protected] (G.G.)

R E P O R T S


addition to the previously identified gene for thesecreted triacylglycerol lipase (PPA2105), vari-ous other lipase/esterase genes can be found inthe genome. Three of these possess a C-terminalLeu-Pro-X-Thr-Gly (LPXTG)–type cell-wallsorting signal. As previously known, a hyaluro-nate lyase (PPA380) degrades hyaluronan, animportant constituent of the extracellular matrixof connective tissues, and presumably aids bac-terial invasion (13, 14). Genome sequencing re-vealed numerous additional enzymes putativelyinvolved in host tissue degradation, such astwo endoglycoceramidases (PPA644 andPPA2106) and four sialidases, two of them pos-sessing a LPXTG motif (PPA1560 andPPA1569). Other degradative enzymes in-clude a putative endo-�-N-acetylglucosamini-dase (PPA990) and various extracellular pepti-

dases. Five highly similar genes encoding ho-mologs to CAMP (Christie, Atkins, Munch-Peterson) factors are also present in the genomeof P. acnes (fig. S4). These factors are secretedproteins characteristic of some streptococcal spe-cies. CAMP factors have been shown to bind toimmunoglobulins of the G and M classes andhave long been known as pathogenic determi-nants. Recently, it was reported that CAMP fac-tors can act as pore-forming toxins (15). Theseproteins could explain previously observed cy-totoxic effects of P. acnes strains (16).

In several studies, the capability of P. acnesto interact with and stimulate the immune systemhas been investigated. Increased cellular, as wellas humoral, immunity to P. acnes has beendetected in patients with severe acne (3–5). Thegenome sequence encodes many factors with

antigenic potential, for example, cell surface pro-teins that may also exhibit cell-adherent proper-ties (Table 1 and fig. S3). Several of these(PPA1879-1881, PPA1955, PPA2127, andPPA2210) possess a C-terminal LPXTG-typecell-wall sorting signal that is required for attach-ing surface proteins to the cell-wall through theaction of sortase (possibly encoded by candidate:PPA777), a mechanism employed by manyGram-positive bacteria (17). In all, 25 genesencoding proteins with a C-terminal LPXTGmotif could be found in the genome.A few of these (PPA1880, PPA2127, andPPA2130) possess contiguous stretches of gua-nidine (G) or cytosine (C) residues, either in theputative promoter region or within the 5� end.The sequences of several plasmids (of the shot-gun library) that cover these regions were am-

Table 1. Selected factors of P. acnes that are putatively involved in degrading host molecules, conferring cell adhesion and/or mediating inflammation. ORF,open reading frame.

ORF number(s) Function Comment Homology

380 Hyaluronate lyase Polysaccharide lyase familyprotein

94% identity to P. acnes clone49/51

570, 1035, 1101, 1224, 1425,1631, 1745, 1761, 1839,1953, 1967, 2036, 2142,2150, etc.

Putative lipases/esterases ORFs 570, 1745, and 2150have a C-terminal LPXTGmotif

Diverse

644, 2106 Endoglycoceramidases Hydrolyzes glycosphingolipids Rhodococcus sp., Cyanea nozakii684 Sialidase L Trans-sialidase Macrobdella decora (leech)685 Sialidase A Exo-�-sialidase Clostridium perfringens, C.

septicum687, 1198, 1231, 1340, 2108 CAMP factors Immunoglobulin-binding,

pore-forming toxinStreptococcus species

990 Endo-�-N-acetylglucosaminidase Streptomyces plicatus1396 Putative hemolysin M. ulcerans, M. leprae1560, 1569 Sialidases/neuraminidases With LPXTG motif Micromonospora viridifaciens1796, 2105 Triacylglycerol lipases ORF 2105 is 100% identical to

gehAP. acnes P-37

1819–1821 Putative chitobiase/�-N-acetylhexosaminidase

Possible frameshift, withLPXTG motif

Arthrobacter

2139 Putative cutinase Botryotinia fuckeliana109 Myosin-crossreactive antigen Rhodopseudomonas palustris125–134, 145–150, 1692–1700,

1185, 1791, 2181Glycosyltransferases, uridine

diphosphate–N-acetylglucosamine2-epimerase, polysaccharidebiosynthesis proteins

Slime/capsular polysaccharidebiosynthesis

Staphylococcus aureus and others

453, 1772, 1773916, 203820392040, 1098737

GroEL, Cpn10DnaJ2, DnaJGrpEDnaK18-kD antigen

The homolog of ORF 737 is amajor immune reactiveprotein in mycobacteria

100% identity to GroEL and DnaKof P. acnes strain P-37

721, 1962 Homologs of invasion-associatedprotein p60

NlpC/P60 family Listeria welshimeri

571 Putative secreted surface protein WD domain, senescencemarker protein-30 domain

765 Immunogenic protein, antigen84

divIVA domain Corynebacterium efficiens, M.leprae, M. tuberculosis

1983, 1984, 1906–1908,1663–1666

Putative adhesions Thrombospondin type 3repeats, PKD and SESTdomains

1955 Surface-associated protein RTX toxin domain, LPXTGmotif

1879, 1880, 1881, 2127 PTRPs Putatively regulated by phasevariation, LPXTG motifs

1715, 2210, 2270 PTRPs ORF 2210 has a LPXTG motif2130 Outer membrane protein A

family proteinPutatively regulated by phase

variation

R E P O R T S


biguous with respect to the length of the poly(C)/(G) stretches (supporting online text and figs. S5and S6). Such variable homopolymeric C or Gstretches, generated in the course of replicationbecause of slipped-strand mispairing, have beenreported in other species to be involved in phasevariation, an adaptation strategy to generatephenotypic variation (18) (supporting onlinetext). Although showing no similarity todatabase entries, PPA1880 and PPA2127both contain characteristic multiple repeatsof the dipeptide proline-threonine [PT repet-itive protein (PTRP)]. Such PT repeats havebeen detected in antigenic proteins of My-cobacterium tuberculosis (19). AdditionalPTRPs (PPA1715, PPA2210, and PPA2270)were found in P. acnes with characteristicsof surface proteins, which may representfurther host-interacting factors (Table 1 andsupporting online text).

Several HSPs have been identified as majortargets of the immune response in bacterialpathogens. Homologs of GroEL and DnaKwere found and characterized in P. acnes (6).The genome sequence contains several moreHSPs, such as DnaJ (PPA916 and PPA2038),GrpE (PPA2039), and an 18-kD protein(PPA737), the homologs of which in my-cobacteria are major immune reactive proteins(20). Further proteins show similarities toknown bacterial immunogenic factors, such asPPA765, to antigen 84 of M. tuberculosis andM. leprae, which is a highly immunogenic pro-tein involved in the symptoms of multibacillaryleprosy (21). Porphyrins are produced by P.acnes in high amounts, and these are alsothought to be involved in inflammation (1)(supporting online text). In the presence of in-creasing oxygen tension, the interaction of mo-lecular oxygen with released porphyrins gener-ates toxic, reduced oxygen species, which candamage keratinocytes and lead to cytokine re-lease (fig. S7).

In summary, the genome sequence clearlyreveals many proteins involved in the ability ofP. acnes to colonize and reside in human skinsites as well as a pronounced potential to sur-vive a spectrum of environments. This capacityhelps to explain the ubiquity of P. acnes andalso its potential hazards, for example, thepublic health problems associated with BloodBank contaminations, or the contamination ofthe human genome sequence database withP. acnes sequence. The GenBank entryAAH14236.1, a proposed human protein, is infact a P. acnes protein (PPA1069).

References and Notes1. K. T. Holland et al., Dermatology 196, 67 (1998).2. J. E. Miskin, A. M. Farrell, W. J. Cunliffe, K. T. Holland,

Microbiology 143, 1745 (1997).3. E. Ingham, Curr. Opin. Infect. Dis. 12, 191 (1999).4. A. Koreck, A. Pivarcsi, A. Dobozy, L. Kemeny, Derma-

tology 206, 96 (2003).5. U. Jappe, E. Ingham, J. Henwood, K. T. Holland, Br. J.

Dermatol. 146, 202 (2002).

6. M. D. Farrar, E. Ingham, K. T. Holland, FEMS Microbiol.Lett. 191, 183 (2000).

7. E. Jakab et al., Yale J. Biol. Med. 69, 477 (1996).8. T. Yamada et al., J. Pathol. 198, 541 (2002).9. The sequence reported in this paper has been depos-

ited in GenBank with accession no. AE017283.10. Materials and methods are available as supporting

material on Science Online.11. S. Hujer, P. Durre, unpublished data.12. E. M. Gribbon, W. J. Cunliffe, K. T. Holland, J. Gen.

Microbiol. 139, 1745 (1993).13. B. Steiner, S. Romero-Steiner, D. Cruce, R. George,

Can. J. Microbiol. 43, 315 (1997).14. G. Makris, J. D. Wright, E. Ingham, K. T. Holland,

Microbiology 150, 2005 (2004).15. S. Lang, M. Palmer, J. Biol. Chem. 278, 38167 (2003).16. Z. Csukas, B. Banizs, F. Rozgonyi, Microb. Pathog. 36,

171 (2004).17. D. Comfort, R. T. Clubb, Infect. Immun. 72, 2710 (2004).18. A. van der Ende et al., J. Bacteriol. 177, 2475 (1995).19. K. K. Singh, X. Zhang, A. S. Patibandla, P. Chien, S. Laal,

Infect. Immun. 69, 4185 (2001).

20. R. J. Booth et al., Infect. Immun. 61, 1509 (1993).21. P. W. Hermans et al., Infect. Immun. 63, 954 (1995).22. This project was carried out within the framework of

the Competence Network Gottingen “Genome Re-search on Bacteria” (GenoMik), financed by the Ger-man Federal Ministry of Education and Research(BMBF). A supporting grant was provided by theNiedersachisches Ministerium fur Wissenschaft undKultur to the Gottingen Genomics Laboratory. H.B. isholder of a fellowship of the German Academy ofNatural Scientists Leopoldina, funded by the BMBF.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/671/DC1Materials and MethodsSOM TextFigs. S1 to S7Table S1References and Notes

14 May 2004; accepted 30 June 2004

Synthetic Mammalian PrionsGiuseppe Legname,1,2* Ilia V. Baskakov,1,2*† Hoang-Oanh B. Nguyen,1

Detlev Riesner,6 Fred E. Cohen,1,3,4 Stephen J. DeArmond,1,5

Stanley B. Prusiner1,2,4‡

Recombinant mouse prion protein (recMoPrP) produced in Escherichia coli waspolymerized into amyloid fibrils that represent a subset of � sheet–rich struc-tures. Fibrils consisting of recMoPrP(89–230) were inoculated intracerebrallyinto transgenic (Tg) mice expressing MoPrP(89–231). The mice developedneurologic dysfunction between 380 and 660 days after inoculation. Brainextracts showed protease-resistant PrP by Western blotting; these extractstransmitted disease towild-type FVBmice and Tgmice overexpressing PrP, withincubation times of 150 and 90 days, respectively. Neuropathological findingssuggest that a novel prion strain was created. Our results provide compellingevidence that prions are infectious proteins.

Prion diseases are responsible for some dev-astating neurological diseases, includingCreutzfeldt-Jakob disease in humans, andpresent as infectious, genetic, and sporadicillnesses (1). The production of a new prionstrain in Tg mice expressing an artificial,chimeric PrP transgene (2) encouraged us torenew our effort to produce synthetic wild-type prions. Earlier, we and others were un-able to produce prion infectivity with the useof recombinant wild-type PrP refolded into �sheet–rich isoforms (3, 4); hence, we turnedto mutant PrPs.

Mice expressing high levels of MoPrP-(P101L), which harbors a mutation (Pro101

3 Leu) analogous to the human prion protein

mutation that causes Gerstmann-Straussler-Scheinker syndrome, develop neurodegen-eration spontaneously at an early age. Brainextracts prepared from these Tg mice trans-mit prion disease to Tg mice expressing lowlevels of MoPrP(P101L), designated Tg196mice (5). About 30% of Tg196 mice developspontaneous illness at �550 days of age. Wesynthesized a 55–amino acid peptide composedof MoPrP residues 89 to 143 with the P101Lmutation and folded it into a �-rich conforma-tion. The aggregated peptide produced neuro-logic dysfunction in Tg196 mice within �1year after inoculation, whereas the non–�-richform did not (6). The incubation time for thesemutant prions did not change upon serial pas-sage (7). Even though the disease-causingMoPrPSc(P101L) readily formed amyloid in thebrains of Tg196 mice, resistance of the proteinto limited proteolysis could be demonstratedonly under “mild” digestion conditions.

Although PrP amyloid deposition in brain ispathognomonic of prion disease, it is a nonoblig-atory feature (8). Moreover, full-length PrPSc

does not polymerize into amyloid, whereas N-terminally truncated PrPSc (designated PrP 27-30) assembles into amyloid fibrils (9). On thebasis of the foregoing findings, we reasoned that

1Institute for Neurodegenerative Diseases, Depart-ments of 2Neurology, 3Cellular and Molecular Phar-macology, 4Biochemistry and Biophysics, and 5Pathol-ogy, University of California, San Francisco, CA 94143,USA. 6Institut fur Physikalische Biologie, Heinrich-Heine-Universitat, 40225 Dusseldorf, Germany.

*These two authors contributed equally to this work.†Present address: Medical Biotechnology Center, Uni-versity of Maryland Biotechnology Institute, Balti-more, MD 21201, USA.‡To whom correspondence should be addressed. E-mail: [email protected]

R E P O R T S


PrP amyloid represents a limited subset of �-richPrPs, one or more of which might be infectious.Two protocols were used to produce the fibrilsfrom recMoPrP(89–230) expressed in E. coli(10) (fig. S1); one used monomeric recMo-PrP(89–230) to produce the amyloid fibrils (de-noted “unseeded”), and the other used some ofthe unseeded fibrils as a seed for the productionof nascent fibrils (denoted “seeded”) (11).

The seeded and unseeded amyloid fibrilscomposed of recMoPrP(89–230) were inocu-lated intracerebrally into Tg(MoPrP,‚23–88)9949/Prnp0/0 mice (henceforth Tg9949mice) expressing MoPrP(89–231) at a level 16times that of normal PrPC in Syrian hamster(SHa) brain. In Tg9949 mice, we do not knowwhether the glycolipid anchor of MoPrP(89–231) is attached to Ser230 or Ser231. The twoamyloid preparations as well as control phos-phate-buffered saline (PBS) were prepared andinjected on the same day. If contamination hadbeen a problem, then the PBS-inoculated miceshould have developed disease (table S1) unlessthe amyloids were selectively tainted.

Mice receiving recMoPrP(89–230) amy-loid developed neurologic disease between380 and 660 days after inoculation (Fig. 1A)(table S1). In earlier studies, uninoculatedTg9949 mice lived for more than 500 dayswithout any signs of neurologic dysfunction(12). In the study reported here, Tg9949 micewere healthy at �670 days of age and failedto show any signs of disease at 620 days afterinoculation with PBS or 525 days with SHaSc237 prions (table S1).

The shortest incubation times for Tg9949mice inoculated with seeded amyloid andunseeded amyloid were 382 days and 474days, respectively. Western blot analysis ofbrain homogenates of these two mice re-vealed that the Tg9949 mouse inoculatedwith seeded amyloid had more protease-resistant PrPSc than the brain of the unseeded

amyloid-inoculated mouse (Fig. 1B). Wheth-er the different incubation times and diversebiochemical profiles reflect higher levels ofPrPSc in the seeded amyloid relative to theunseeded amyloid, or whether these findingssignify the creation of two different prionstrains, remains to be established. No pro-tease-resistant PrP was found by Westernblotting in the brain of a Tg9949 mouseinoculated with PBS and killed at 530 daysafter inoculation; by this time, five of sevenTg9949 mice inoculated with seeded recMo-PrP(89–230) amyloid had succumbed to dis-ease (Fig. 1B). Consistent with the view thatthe amyloid fibrils contain low levels of Mo-PrPSc(89–230), we were unable to detect thisprotein in fibril preparations by immunoblot-ting, and we observed incubation times of�500 days on first passage followed byshortening to �250 days on second passage(table S1). For all strains of prions studied todate, low titer inocula produce prolonged in-cubation times that shorten upon subsequentpassage (13). On first passage, mice do notbecome ill until the prion titers in their brainsreach maximal levels. Whether the amyloidfibrils protected the small amounts of PrPSc

found within them or modified the retentionof PrPSc in brain after inoculation remains tobe determined.

Residual MoPrPC(89–231) after protein-ase K (PK) digestion of brain homogenatefrom a PBS-inoculated Tg9949 mouse re-flects the high level of PrPC expression (Fig.1B, lane 5), a problem encountered withsome other Tg mice. Fortunately, the level ofresidual MoPrPC(89–231) after PK digestionis sufficiently low as to not present problemswith the interpretation of the data.

Tg9949 mice inoculated with seeded amy-loid exhibited extensive vacuolation with gli-osis in the cerebellum, hippocampus, brain-stem, and white matter (fig. S2A). The

distribution, density, and morphology of thevacuoles were different for unseeded andseeded amyloid, raising the possibility thatthey represent two different prion strains (fig.S2A). Vacuolation, astrocytic gliosis, andPrPSc accumulation were more widely dis-persed in gray matter regions in the brains ofmice inoculated with unseeded amyloid rela-tive to seeded amyloid (Fig. 2, A to D). Theneuroanatomic distributions of vacuoles asso-ciated with unseeded and seeded amyloidwere different from those found with mouseRML prions (compare fig. S2A with fig.S2B). With unseeded amyloid, the majorityof vacuoles measured 20 to 50 �m in diam-eter (Fig. 2A), whereas most vacuoles fromRML prions were 10 to 30 �m in diameter(Fig. 2E). From the seeded amyloid inocu-lum, smaller (10 to 20 �m) and larger (20 to50 �m) vacuoles were evenly represented(Fig. 2C). With both unseeded and seededamyloid, PrPSc was deposited in gray matteras relatively large solitary masses 5 to 20 �min diameter and at the perimeter of manyvacuoles (Fig. 2, B and D). In contrast, RML-infected mice exhibited fine granular accu-mulations of PrPSc (Fig. 2F).

Prions in the brains of Tg9949 mice inoc-ulated with seeded amyloid were designatedMoSP1 (mouse synthetic prion strain 1). Serialtransmission of MoSP1 prions from Tg9949mice to wild-type FVB mice and to Tg(MoPrP-A)4053 mice (henceforth Tg4053 mice) gavemean incubation times of 154 and 90 days,respectively (Fig. 3) (table S1). Tg4053 miceexpress MoPrP-A at a level 8 times that ofSHaPrPC in hamster brain (14). Protease-resistant PrPSc was found in the brains of bothwild-type FVB and Tg4053 mice inoculatedwith MoSP1 prions (Fig. 1C).

Well-defined PrP amyloid plaques as wellas numerous, densely packed, finely granularPrPSc deposits were identified in the brains of

Fig. 1. Survival curves and immunoblots of Tg9949 mice. (A) Tg9949 mice inoculated with RML prions(■ ), seeded-amyloid recMoPrP(89–230) (Œ), and unseeded-amyloid recMoPrP(89–230) (‚). Amyloidfibrils were formed upon incubation of recMoPrP(89–230) (0.6 mg/ml) at 37°C in 3 M urea, 0.2 MNaCl, 50 mM sodium acetate buffer, pH 5.0, as described (11). The kinetics of fibril formation were

monitored with a thioflavin T binding assay (24). Inocula were prepared by dialysis of the fibrils in PBS buffer, pH 7.2, for 2 days. Concentration ofrecMoPrP(89–230) in the inocula was �0.5 mg/ml. (B) Immunoblot of PrPSc in brains of Tg9949 mice. Paired-samples lanes: 1, uninoculated normalCD1 mouse; 2, RML-inoculated CD1 mouse; 3, Tg9949 mouse inoculated with seeded amyloid; 4, Tg9949 mouse inoculated with unseeded amyloid;5, Tg9949 mouse inoculated with PBS and killed at 580 days of age; 6, Tg9949 mouse inoculated with RML prions. (C) Immunoblot of PrPSc in brainsof wild-type CD1, wild-type FVB, and Tg4053 mice. Paired-samples lanes: 1, uninoculated normal CD1 mouse; 2, RML-inoculated CD1 mouse; 3,RML-inoculated FVB mouse; 4 and 5, Tg4053 mouse inoculated with MoSP1 (brain homogenate from Tg9949 mice inoculated with seeded recMoPrPamyloid); 6 and 7, FVB mouse inoculated with MoSP1 (brain homogenate from Tg9949 mice inoculated with seeded recMoPrP amyloid). Symbols in(B) and (C): –, undigested control sample; �, samples subjected to limited PK digestion. Apparent molecular weights based on migration of proteinstandards are indicated.

R E P O R T S


both FVB (Fig. 4F) and Tg4053 mice inocu-lated with MoSP1 prions (15). The vacuola-tion scores (the percentage of an area occu-pied by vacuoles) were greater in the brainsof Tg4053 mice than in those of FVB mice(fig. S3A). Additionally, the density of vacu-oles (number per area) seen with MoSP1prions was greater than that seen with RMLprions in Tg4053 mice (fig. S3B). RML pri-ons failed to cause vacuolation in the caudatenucleus, septal nuclei, and cerebellar whitematter in Tg4053 mice. These findings arguethat the MoSP1 strain remained stable duringpassage from Tg9949 to Tg4053 mice butseemed less stable upon passage in FVBmice, consistent with the pattern of PrP im-munostaining (Fig. 4F).

The specificity of the response in Tg9949mice to recMoPrP(89–230) polymerized intoamyloid was established by the absence ofdisease after inoculation with PBS and Sc237prions (table S1). When the healthy Tg9949mice inoculated with Sc237 prions werekilled at 525 days after inoculation, five ofseven Tg9949 mice inoculated with the seed-ed amyloid were already ill.

Many attempts to solubilize PrPSc undernondenaturing conditions and to renature PrP27-30 have been unsuccessful (16, 17);hence, progress in elucidating the structure ofPrPSc has been slow (18). In purified frac-tions, PrPSc forms amorphous aggregates,whereas PrP 27-30 spontaneously assemblesinto amyloid polymers in the presence ofdetergent (19). Some investigators have ar-gued that prions are infectious amyloids, ig-noring findings that full-length PrPSc doesnot form amyloid (20, 21). Notably, thefibrillar morphology of prion rods composedof PrP 27-30 disappeared after exposure tothe organic solvent 1,1,1-trifluoro-2-propanol( TFIP) while prion infectivity was retained(22). Also of interest is a short peptide com-posed of 27 amino acids corresponding to PrPresidues 100 to 126 that readily forms amy-loid (23). On this background, we pursuedinvestigations directed at producing syntheticprions using PrP amyloid as a surrogatemarker for the folding of MoPrP(89–230)into a biologically active conformation. Therapidity and ease of measuring thioflavin T

binding that reflects amyloid formation (24)permitted us to determine conditions underwhich recMoPrP(89–230) would polymerize.

Our finding that amyloid fibrils harbor de-tectable levels of prion infectivity allows us todraw some conclusions about mammalian pri-ons. First, PrP is both necessary and sufficientfor infectivity. Second, neither the Asn-linkedoligosaccharides nor the glycosylphosphatidy-linositol anchor are required for prion infectiv-ity, because recMoPrP(89–230) contains nei-ther of these posttranslational modifications(25–28). Whether the Asn-linked oligosaccha-rides or the glycosylphosphatidylinositol an-chor increase the efficiency of PrPSc formationfor particular prion strains is unknown (29).Third, variations in PrP glycosylation are notrequired for prion diversity, and the biolog-ical information carried by distinct strainsof prions resides in PrPSc, as previouslysuggested (2, 30, 31). Fourth, spontaneousformation of prions, which is thought to beresponsible for sporadic forms of prion dis-ease in livestock and humans, can occur inany mammal expressing PrPC (32).

Fig. 2. The distinguishing neuropathologicalfeatures of unseeded recPrP amyloid (A and B),seeded recPrP amyloid (C and D), and RMLprions (E and F) in the pons of Tg9949 mice.Left column: hematoxylin and eosin (H&E) stain(scale bar, 50 �m); right column: immunohisto-chemistry of PrPSc by the hydrated autoclavingmethod using the PrP-specific HuM-R2 monoclo-nal antibody fragment (43) (scale bar, 25 �m).

Fig. 3. (A) Survivalcurves of FVB mice in-oculated with RML (‚)and MoSP1 (}) prions.(B) Survival curves ofTg4053 mice inoculat-ed with RML (‚) andMoSP1 (}) prions.

Fig. 4. Comparison of neuropathological changes in the pons associated with inoculation of seededrecPrP preparations into Tg9949 mice (A to C) and of MoSP1 (prions derived from clinically illTg9949 mice inoculated with seeded recPrP amyloid) inoculated into FVB mice (D to F). Bothpassages show the neurohistological characteristics of a prion disease: vacuoles (spongiformdegeneration) shown by H&E staining (left); reactive astrocytic gliosis shown by glial fibrillaryacidic protein immunohistochemistry (center); and accumulation of PrPSc visualized by hydratedautoclaving immunohistochemistry with HuM-R2 monoclonal antibody (right). Scale bars, 50 �m[(A), (B), (D), (E)], 25 �m [(C) and (F)].

R E P O R T S


Although we have used the polymerizationof recMoPrP(89–230) into amyloid fibrils togenerate prion infectivity, we hasten to add thatother �-rich forms of recMoPrP(89–230) mayalso harbor infectivity. Preliminary results sug-gest that preparations of �-oligomers formedfrom recMoPrP(89–230) may also contain lowlevels of prion infectivity (33). Such findingsemphasize the need to define optimal conditionsfor prion formation in vitro under which highlevels of PrPSc can be generated. Moreover, pre-vious difficulties in creating infectious prions invitro from recPrPs enriched for �-structure maybe due to the tendency of mammalian PrPs tofold into biologically irrelevant �-rich isoforms(3, 4, 11). In studies of fungal prions, the ease ofassaying infectivity (34) and the ability to studymillions of colonies made the creation of in vitroinfectivity from recombinant proteins more trac-table (35–37). Whereas yeast prions form withinthe cytoplasm (38), mammalian prions arethought to be produced on the cell surface incaveolae-like domains (39, 40).

From Tg mouse studies, it is well establishedthat templates improve the likelihood of formingan infectious �-rich isoform (8, 12). In the stud-ies reported here, we see evidence that seededamyloid fibrils exhibit shorter incubation timesthan their unseeded progenitor (Fig. 1A). It re-mains to be determined whether this is due to thegreater number of PrPSc molecules within seededfibrils relative to unseeded fibrils, or whether thisreflects strain differences.

Our results have important implications forhuman health. The formation of prions fromrecPrP demonstrates that PrPC is sufficient forthe spontaneous formation of prions; thus, noexogenous agent is required for prions to form inany mammal. Our findings provide an explana-tion for sporadic Creutzfeldt-Jakob disease forwhich the spontaneous formation of prions hasbeen hypothesized. Understanding sporadicprion disease is particularly relevant to control-ling the exposure of humans to bovine prions(41). That bovine prions are pathogenic for hu-mans is well documented; more than 150 teen-agers and young adults have already died fromprion-tainted beef derived from cattle with bo-vine spongiform encephalopathy (42). More-over, the sporadic forms of Alzheimer’s andParkinson’s diseases as well as amyotrophic lat-eral sclerosis and the frontal temporal dementiasare the most frequent forms of these age-dependent disorders, as is the case for the priondiseases. Important insights in the etiologicevents that feature in these more commonneurodegenerative disorders, all of which arecaused by the aberrant processing of proteinsin the nervous system, are likely to emerge asmore is learned about the molecular pathogen-esis of the sporadic prion diseases.

References and Notes1. S. B. Prusiner, in Prion Biology and Diseases, S. B.

Prusiner, Ed. (Cold Spring Harbor Laboratory Press,Cold Spring Harbor, NY, 2004), pp. 89–141.

2. D. Peretz et al., Neuron 34, 921 (2002).3. I. Mehlhorn et al., Biochemistry 35, 5528 (1996).4. A. F. Hill, M. Antoniou, J. Collinge, J. Gen. Virol. 80, 11

(1999).5. K. K. Hsiao et al., Proc. Natl. Acad. Sci. U.S.A. 91,

9126 (1994).6. K. Kaneko et al., J. Mol. Biol. 295, 997 (2000).7. P. Tremblay et al., J. Virol. 78, 2088 (2004).8. S. B. Prusiner et al., Cell 63, 673 (1990).9. S. B. Prusiner et al., Cell 35, 349 (1983).

10. See supporting data on Science Online.11. I. V. Baskakov, G. Legname, M. A. Baldwin, S. B.

Prusiner, F. E. Cohen, J. Biol. Chem. 277, 21140(2002).

12. S. Supattapone et al., J. Virol. 75, 1408 (2001).13. S. B. Prusiner, M. R. Scott, S. J. DeArmond, G. Carlson,

in Prion Biology and Diseases, S. B. Prusiner, Ed. (ColdSpring Harbor Laboratory Press, Cold Spring Harbor,NY, 2004), pp. 187–242.

14. G. C. Telling et al., Genes Dev. 10, 1736 (1996).15. G. Legname et al., data not shown.16. K. Post, D. R. Brown, M. Groschup, H. A. Kretzschmar,

D. Riesner, Arch. Virol. Suppl., 265 (2000).17. H. Wille, S. B. Prusiner, Biophys. J. 76, 1048 (1999).18. C. Govaerts, H. Wille, S. B. Prusiner, F. E. Cohen, Proc.

Natl. Acad. Sci. U.S.A. 101, 8342 (2004).19. M. P. McKinley et al., J. Virol. 65, 1340 (1991).20. D. C. Gajdusek, Br. Med. Bull. 49, 913 (1993).21. J. T. Jarrett, P. T. Lansbury, Jr., Cell 73, 1055 (1993).22. H. Wille, G.-F. Zhang, M. A. Baldwin, F. E. Cohen, S. B.

Prusiner, J. Mol. Biol. 259, 608 (1996).23. G. Forloni et al., Nature 362, 543 (1993).24. H. LeVine, Protein Sci. 2, 404 (1993).25. A. Taraboulos et al., Proc. Natl. Acad. Sci. U.S.A. 87,

8262 (1990).26. S. J. DeArmond et al., Neuron 19, 1337 (1997).27. P. Gambetti, P. Parchi, N. Engl. J. Med. 340, 1675

(1999).28. J. A. Mastrianni et al., N. Engl. J. Med. 340, 1630

(1999).29. J. Collinge, K. C. L. Sidle, J. Meads, J. Ironside, A. F. Hill,

Nature 383, 685 (1996).

30. R. A. Bessen, R. F. Marsh, J. Virol. 68, 7859 (1994).31. G. C. Telling et al., Science 274, 2079 (1996).32. S. B. Prusiner, Annu. Rev. Microbiol. 43, 345 (1989).33. G. Legname et al., unpublished data.34. R. B. Wickner, Science 264, 566 (1994).35. M. L. Maddelein, S. Dos Reis, S. Duvezin-Caubet, B.

Coulary-Salin, S. J. Saupe, Proc. Natl. Acad. Sci. U.S.A.99, 7402 (2002).

36. C. Y. King, R. Diaz-Avalos, Nature 428, 319 (2004).37. M. Tanaka, P. Chien, N. Naber, R. Cooke, J. S. Weiss-

man, Nature 428, 323 (2004).38. M. M. Patino, J.-J. Liu, J. R. Glover, S. Lindquist, Science

273, 622 (1996).39. A. Gorodinsky, D. A. Harris, J. Cell Biol. 129, 619

(1995).40. P. J. Peters et al., J. Cell Biol. 162, 703 (2003).41. C. Casalone et al., Proc. Natl. Acad. Sci. U.S.A. 101,

3065 (2004).42. R. G. Will, M. P. Alpers, D. Dormont, L. B. Schonberger,

in Prion Biology and Diseases, S. B. Prusiner, Ed. (ColdSpring Harbor Laboratory Press, Cold Spring Harbor,NY, 2004), pp. 629–671.

43. D. Peretz et al., Nature 412, 739 (2001).44. We thank P. Culhane, K. Giles, P. Lessard, M. Scott, S.

Supattapone, and the staff at the Hunters PointAnimal Facility. Supported by NIH grants AG02132,AG10770, and AG021601, the G. Harold and Leila Y.Mathers Charitable Foundation, the Dana Founda-tion, and the Sherman Fairchild Foundation. G.L., D.R.,F.E.C., S.J.D., and S.B.P. have a financial interest inInPro Biotechnology, which applies research on prionsand prion diseases to promote human health.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/673/DC1Materials and MethodsFigs. S1 to S3Table S1References

11 May 2004; accepted 1 July 2004

Host-to-Parasite Gene Transfer inFlowering Plants: PhylogeneticEvidence from Malpighiales

Charles C. Davis1* and Kenneth J. Wurdack2

Horizontal gene transfer (HGT) between sexually unrelated species has recentlybeen documented for higher plants, butmechanistic explanations for HGTs haveremained speculative. We show that a parasitic relationship may facilitate HGTbetween flowering plants. The endophytic parasites Rafflesiaceae are placed inthe diverse order Malpighiales. Our multigene phylogenetic analyses of Mal-pighiales show that mitochrodrial (matR) and nuclear loci (18S ribosomal DNAand PHYC) place Rafflesiaceae in Malpighiales, perhaps near Ochnaceae/Clusiaceae. Mitochondrial nad1B-C, however, groups them within Vitaceae,near their obligate host Tetrastigma. These discordant phylogenetic hypothesesstrongly suggest that part of the mitochondrial genome in Rafflesiaceae wasacquired via HGT from their hosts.

Malpighiales are one of the most diverseclades of flowering plants uncovered in re-cent phylogenetic analyses. The order com-prises 27 families (1) previously assigned to13 different orders (2), including more than16,000 species spanning tremendous mor-phological and ecological diversity (3). Re-cent surprising additions to Malpighiales arethe endophytic holoparasites Rafflesiaceae(4), which lack leaves, stems, and roots, and

rely entirely on their host plants, species ofTetrastigma (Vitaceae), for their nutrition.Despite their extreme vegetative reduction,they are unmistakable in flower, producingthe largest flowers in the world, which mimicrotting flesh—an enticement to the carrionflies that pollinate them (5).

Barkman et al. (4 ) used mitochondrial(mt) matR sequences to place Raffle-siaceae firmly with Malpighiales [100%

R E P O R T S


bootstrap percentage (BP)]. Their use of asingle mt gene was appropriate in a familythat has resisted placement with standardgenetic loci. To further examine this place-ment, we obtained sequences representingall families of Malpighiales, all genera ofRafflesiaceae, and numerous basal eudi-cots for four loci from the mt and nucleargenomes (6 ). Low-copy nuclear genes arean underused resource for resolving theplacement of problematic taxa, and phyto-chrome C (PHYC), as used here, has beenuseful for revealing relationships withinMalpighiales (7 ).

Our phylogenetic analyses are summa-rized in Fig. 1 (8). The tree created from the

matR and nuclear loci firmly (100% BP)place Rafflesiaceae within Malpighiales. Incontrast, the mt locus nad1B-C suggests thatRafflesiacecae are not members of Mal-pighiales but belong (100% BP) in Vitaceaenear their host Tetrastigma. Each of thesemutually exclusive hypotheses cannot be at-tributable to contamination (9), and each re-ceives strong support from parsimony analy-ses and from alternative topology tests.

Which of these conflicting hypothesesreflect the true species affinities of Rafflesi-aceae? Vitaceae possess several synapomor-phies that are rare among angiosperms, includ-ing sieve-tube plastids with starch and proteininclusions, pearl glands, stamens opposite thepetals, and seeds with a cordlike raphe. IfRafflesiaceae were embedded in Vitaceae, assuggested by nad1B-C, we would expect spe-cies to possess at least some of these characters,but they do not (2, 3). A definitive malpighi-alean sister group for Rafflesiaceae is unclear,given our data. However, the closest relativessuggested in the combined analysis (10),

Ochnaceae and Clusiaceae sensu lato, sharetenuinucellate ovules (among mostly crassinu-cellate relatives) and staminal fusion withRafflesiaceae (2, 3).

The position of Rafflesiaceae based onnad1B-C provides a new example of horizon-tal gene transfer. If nad1B-C were verticallytransmitted, as we believe to be the case forthe other loci, we would expect Rafflesiaceaeto group with Malpighiales. Instead, phylo-genetic evidence from nad1B-C suggests thatpart of the mt genome in Rafflesiaceae orig-inated from their hosts, Tetrastigma (eitherstem or crown group members), and washorizontally transferred to these obligate par-asites. A similar horizontal gene transfer(HGT) of nad1B-C was recently reported (11)in seed plants, involving a transfer from anasterid to Gnetum. And Bergthorsson et al. (12)have documented several instances of mt HGTbetween distantly related angiosperm groups.

The underlying mechanism for HGT be-tween sexually unrelated plants, however, hasbeen elusive. Various pathogens have beensuggested as primary vector agents (11, 12).Our study documents a case in which there isno need to propose an intermediary vector forHGT. In these plants, the transfer appears tohave been facilitated by the intimacy of theassociation between the host and the endo-phytic parasite, which lives its whole vegeta-tive life as “an almost mycelial haustorialsystem,” “ramifying and anastomosingthroughout the [tissues of the] host” (13).This pattern may be an important mechanismby which parasites assemble their geneticarchitecture, and additional cases of HGTshould be sought among other endophyticparasites and their hosts. It will also not besurprising if reciprocal genetic transfers arefound to have occurred, from parasite to host.

References and Notes1. The Angiosperm Phylogeny Group, Bot. J. Linn. Soc.

141, 399 (2003).2. A. Cronquist, An Integrated System of Classification of

Flowering Plants (Columbia Univ. Press, New York,1981).

3. P. F. Stevens, Angiosperm Phylogeny Website, ed. 4(2001 onward) (www.mobot.org/MOBOT/research/Apweb).

4. T. J. Barkman, S.-K. Lim, K. Mat Salleh, J. Nais, Proc.Natl. Acad. Sci. U.S.A. 101, 787 (2004).

5. R. S. Beaman, P. J. Decker, J. H. Beaman, Am. J. Bot.75, 1148 (1988).

6. For primer and GenBank information see the support-ing online material (SOM) on Science Online.

7. C. C. Davis, M. W. Chase, Am. J. Bot. 91, 262 (2004).8. For detailed information on materials and methods,

data sets, phylogenetic analyses, and complete anno-tated figures see the SOM. We were unable to samplePHYC from all families of Malpighiales because it isprobably absent from some clades (7 ). Hence, weanalyzed these data in combination with 18S toensure that all families were represented for thenuclear genome. We also examined the placement ofRafflesiaceae with 18S in two ways. First, we exclud-ed the most divergent domains, V2 and V4, from theanalysis (14 ) (322 of 1813 base pairs)] across all taxa.Second, we treated these domains as missing data

1Ecology and Evolutionary Biology, University ofMichigan Herbarium, 3600 Varsity Drive, Ann Ar-bor, MI 48108–2287, USA. 2Department of Botanyand Laboratories of Analytical Biology, SmithsonianInstitution, 4210 Silver Hill Road, Suitland, MD20746, USA.

*To whom correspondence should be addressed. E-mail: [email protected].

nad1B-CmatR, PHYC, 18SA BFig. 1. Two conflicting hypothesesabout the phylogenetic placementof Rafflesiaceae. (A) The strictconsensus of 136 angiosperms forcombined mt matR and nuclear(PHYC and ribosomal 18S) datashowing a well-supported (100%BP) Malpighiales clade (in blue),which includes all members of theorder sensu APG II (1) plus Raffle-siaceae (in red; Rafflesia, Rhizan-thes, and Sapria). (B) The strictconsensus of 147 angiospermsfor mt nad1B-C (the nad1 intron2 and part of the adjacent exonsb and c) showing a well-supported(100% BP) Malpighiales clade,which includes all members of theorder except Rafflesiaceae. Raffle-siaceae (Rafflesia and Sapria) arestrongly placed (100% BP) in thebasal eudicot family Vitaceae (inyellow) near their host genus, Tet-rastigma. The dashed line is thehypothesized host/parasite HGT.

R E P O R T S


only for Rafflesiaceae. Both approaches yielded con-gruent topologies, with the latter (shown here) beingmuch better resolved. The spurious placement ofRafflesiaceae as sister to all angiosperms when 18S isused (4 ) may be attributable to high divergence inthese small domains and suggests that 18S may notbe as useless for placing problematic taxa as previ-ously suggested (4 ).

9. We took several precautions to avoid and detectcontamination. First, our results were independentlycorroborated in the laboratories of each author. Sec-ond, Rafflesiaceae data were acquired before startingany work on Vitaceae. Third, if our DNA were cross-contaminated we would not expect such stronglyconflicting results regarding the placement of Raffle-siaceae, given the same DNA. Nor would we expectsuch a high degree of sequence divergence ofnad1B-C between Rafflesiaceae and Tetrastigma (orother Vitaceae we sampled). If contamination oc-curred, we would expect sequences to be nearly

identical to those of other sampled Vitaceace, espe-cially given the relatively low amount of sequencedivergence between all accessions of Vitaceae innad1B-C (0.51 to 0.95% sequence divergence). Notonly is nad1B-C divergent within Rafflesiaceae(6.2%), but it is also highly divergent from otherphylogenetically diverse Vitaceae (15) included here(2.5 to 3.3%).

10. Taxa in the combined analysis of 18S, PHYC, andmatR were included if they were sampled for matRplus at least one of the two nuclear loci. We believethis analysis to represent the best estimate of Mal-pighiales phylogeny. Similar approaches combiningmultiple genes have provided powerful insights intoangiosperm phylogeny where single gene studieshave failed (16 ).

11. H. Won, S. S. Renner, Proc. Natl. Acad. Sci. U.S.A.100, 10824 (2003).

12. U. Bergthorsson, K. L. Adams, B. Thomason, J. D.Palmer, Nature 424, 197 (2003).

13. J. Kuijt, The Biology of Parasitic Flowering Plants(Univ. of California Press, Berkeley, CA, 1969).

14. D. L. Nickrent, E. M. Starr, J. Mol. Evol. 39, 62 (1994).15. M. J. Ingrouille et al., Bot. J. Linn. Soc. 138, 421

(2002).16. Y.-L. Qiu et al., Nature 402, 404 (1999).17. We thank W. Anderson, D. Boufford, Y.-L. Qiu, and P.

Stevens. C.C.D. was supported by grants from theUniversity of Michigan, NSF AToL 0431266, and theMichigan Society of Fellows. This paper is dedicatedto V. Morrison.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/1100671/DC1Materials and MethodsFigs. S1 to S5References

25 May 2004; accepted 6 July 2004Published online 15 July 2004; 10.1126/science.1100671Include this information when citing this paper.

KIF1A Alternately Uses TwoLoops to Bind MicrotubulesRyo Nitta,1 Masahide Kikkawa,1,2 Yasushi Okada,1

Nobutaka Hirokawa1*

Themotor protein kinesin moves alongmicrotubules, driven by adenosine triphos-phate (ATP) hydrolysis. However, it remains unclear how kinesin converts thechemical energy into mechanical movement. We report crystal structures of mo-nomeric kinesin KIF1A with three transition-state analogs: adenylyl imidodiphos-phate (AMP-PNP), adenosine diphosphate (ADP)–vanadate, and ADP-AlFx (alumi-nofluoride complexes). These structures, together with known structures of theADP-bound state and the adenylyl-(�,�-methylene) diphosphate (AMP-PCP)–bound state, show that kinesin uses two microtubule-binding loops in an alter-nating manner to change its interaction with microtubules during the ATP hydro-lysis cycle; loop L11 is extended in the AMP-PNP structure, whereas loop L12 isextended in the ADP structure. ADP-vanadate displays an intermediate structurein which a conformational change in two switch regions causes both loops to beraised from the microtubule, thus actively detaching kinesin.

To move along microtubules, kinesins (1)must alternate rapidly between tightly boundand detached states. How both dimeric (2, 3)and monomeric (4, 5) kinesins achieve thisremains unclear. Because the binding energyin the strong-binding state [10 to 20 kBT (3,4), where kB is the Boltzmann constant and Tis absolute temperature] is too large for rapidspontaneous release, the energy for fast de-tachment of kinesin from the microtubulemust come from a step of the ATP hydrolysiscycle. Large change(s) in free energy areexpected to occur during four steps: ATPbinding, hydrolysis, phosphate release, andADP release. Both conventional kinesin andKIF1A bind tightly to microtubules in thenucleotide-free state and in the ATP-bound

state. In the ADP-bound state, conventionalkinesin is detached from microtubules,whereas KIF1A is partially detached and dif-fuses freely along the microtubule. This isbecause loose binding of ADP-bound KIF1Ais supported by the KIF1 family–specific K-loop at loop L12. A mutant KIF1A that lacksthe K-loop detaches from the microtubule inthe ADP-bound state, and the dissociation

constant markedly varies depending on thetype of bound nucleotide, as is true for con-ventional kinesin (4). For historical reasons,the tightly bound state is called the strong-binding state, and the fully or partially de-tached state is called the weak-binding state.Recent work detected the phosphate releasefrom a mutant kinesin, which stalls before thedetached state (6, 7). This means that detach-ment occurs just at or after the phosphaterelease. Thus, the active process to detachkinesin from the microtubule should occur atthe transition from the strong-binding state tothe weak-binding state.

The active detachment process can be de-tected in KIF1A because of its property ofbinding to the microtubule during adenosinetriphosphatase (ATPase) cycling. The apparentdissociation constant of KIF1A in the presenceof ATP is the weighted average of the equilib-rium dissociation constant of various interme-diate states during the ATPase turnover.Because the dissociation constant is not signif-icantly different between two major inter-mediate states, the AMP-PNP–bound andADP-bound states ( Table 1) (fig. S1) (8), theapparent dissociation constant during theATPase turnover was not expected to be fun-damentally different from these values. How-ever, the apparent dissociation constant in thepresence of 2 mM ATP was twice the expected

1Department of Cell Biology and Anatomy, Universityof Tokyo, Graduate School of Medicine, Hongo,Bunkyo-ku, Tokyo 113-0033, Japan. 2Department ofCell Biology, University of Texas Southwestern Med-ical Center, 5323 Harry Hines Boulevard, Dallas, TX75390, USA.


Table 1. (Apparent) equilibrium dissociation constants (Kd) for microtubules. Kd values are reported asmeans � SEM of at least three independent experiments. Conditions: 2 mM nucleotide or its analog, 50mM imidazole, 5 mM Mg-acetate, 1 mM EGTA, and 50 mM K-acetate, pH 7.4 at 27°C (nd, notdetermined).

NucleotideKd (nM)

Wild type L12† L11‡ L8§

AMP-PNP 4.2 � 1.3 6.0 � 1.4 20.2 � 4.0 25.0 � 6.0ADP 6.8 � 2.5 23.5 � 8.4 12.3 � 4.0 26.5 � 5.0ATP* 10.8 � 1.8 40.5 � 11.8 nd ndADP-AlFx 5.9 � 1.5 7.1 � 1.7 nd ndADP-Vi 21.4 � 4.3 167 � 66 nd nd

*ATP regeneration system was used to maintain ATP/ADP level. †L12: CK1 (4 ). ‡L11: K261A/R264A/K266A. §L8: K161A/R167A/R169A/K183A.

R E P O R T S


value ( Table 1) (fig. S1). Considering that theduration of the intermediate states other thanthe AMP-PNP–bound and ADP-bound states ismuch shorter, this large apparent dissociationconstant implies that KIF1A has another inter-mediate state with a considerably lower affinityto the microtubule, which likely corresponds tothe actively detaching state.

Because this detached state must occurbetween the ATP- and ADP-bound states, weexamined the KIF1A-microtubule interactionin the presence of two ADP-phosphate ana-logs, ADP-vanadate (ADP-Vi) and ADP-AlFx (9). Interestingly, the two analogs haddifferent effects. ADP-Vi markedly loweredthe affinity of KIF1A to the microtubule,whereas ADP-AlFx did not significantlychange the affinity ( Table 1) (fig. S1). Thus,ADP-Vi and ADP-AlFx stop the ATPase cy-cle of KIF1A at two different substates of theADP-phosphate state. The ADP-Vi state like-ly corresponds to the actively detaching state(10). Extremely weak binding in the presenceof ADP-Vi is not specific to KIF1A. Forexample, conventional kinesin and the Dro-sophila kinesin-like proteins ncd and nod alsoshow extremely low affinity for microtubulesin the presence of ADP-Vi, whereas ADP-AlFx increases affinity (11, 12). Because theaffinity of conventional kinesin in the ADP-

AlFx state is closer to the affinity of theATP-bound state than to the ADP-Vi or theADP states (13), ADP-AlFx likely stops theATPase cycle of kinesin at a state betweenthe ATP-bound and ADP-Vi states (Fig. 1).

Thus, biochemical analyses suggest that theaffinity of kinesin-type motors to the microtu-bules undergoes a drastic change during thehydrolysis of ATP into ADP. In the ATP-boundstate and in the early ADP-phosphate state(ADP-AlFx), kinesin has its highest affinity tothe microtubules (the strong-binding state).This state is followed by a transient late ADP-phosphate state (ADP-Vi), at which point kine-sin has its lowest affinity (the actively detachingstate). After phosphate release, kinesin in theADP state has a moderate affinity (the weak-binding state).

To understand the atomic details of thesemechanical states, we solved the crystalstructures of KIF1A in the ADP-phosphate(ADP-AlFx and ADP-Vi) states and com-pared them with known structures of theADP-bound state (1I5S) (14) and the AMP-PCP–bound state (1I6I) (14), as well as withan AMP-PNP–bound form solved here. Asreported for myosin (15), AMP-PCP blocksthe ATPase cycle before an isomerizationstep, which precedes ATP hydrolysis. Thus,the AMP-PCP–bound form would represent

a collision complex or a preisomerizationstate, rather than a prehydrolysis state, assuggested from the lack of interactions be-tween the �-phosphate of AMP-PCP and thetwo highly conserved residues in the switchregions (Ser215 and Gly251) (see below) (fig.S2). In this study, we solved the structure inthe AMP-PNP–bound state in order to exam-ine the prehydrolysis state.

The x-ray crystal structures of the motordomain of KIF1A with ADP-Vi, ADP-AlFx, and AMP-PNP were refined to val-ues of 2.2 Å, 2.6 Å, and 1.9 Å, respectively(see table S1 for details of our crystallo-graphic work). To identify the regions thatchange conformation during ATP hydroly-sis, we superimposed these three structuresand the previously solved ADP-boundstructure with the use of the C� atoms ofthe conserved nucleotide-binding loop (P-loop: amino acids 97 to 104), which re-mains unchanged throughout ATP hydroly-sis (Fig. 2) (14 ). The overall architecturesof the ADP-Vi, ADP-AlFx, and AMP-PNPforms are similar to the previously solvedADP and AMP-PCP forms. However, al-though the central � sheets and surround-ing � helices hardly change their confor-mations [root mean square deviation(RMSD) � 0.6 Å], three loop regions doundergo marked changes (16). These arethe switch I region (loop L9; amino acids202 to 218), the switch II cluster (loop L11,helix �4, loop L12, helix �5, and loop L13;amino acids 248 to 324), and the neck-linker region (strand �9; amino acids 353to 361), which are all thought to play acrucial role in the nucleotide-driven motil-ity of kinesin-type motors.

The switch I loop L9 changes its con-formation, along with the movement andrelease of the �-phosphate, during ATPhydrolysis. Before isomerization, loop L9has a tight �-hairpin structure (AMP-PCPform, Fig. 3A). This structure melts so that,in the prehydrolysis state, the conservedserine residue (Ser215) at the C-terminalend of the loop can approach the �-phosphate (AMP-PNP form) (Fig. 3B) (fig.S2). After hydrolysis, the �-phosphatemoves toward the gap between the switch Iloop L9 and the switch II loop L11 (ADP-AlFx form) (Fig. 3C) (fig. S3) (17), and theloop-to-helix transition of loop L9 likelysupports the release of the �-phosphate outof the nucleotide-binding pocket (ADP-Viform) (Fig. 3D) (fig. S4) (18, 19).

This dynamic structural change in theswitch I loop L9 might explain the rapidhydrolysis and release of the �-phosphate.The �-phosphate is efficiently releasedthrough the “back door” formed by the highlyconserved residues Arg216 and Glu253 (fig.S4). In myosin, the salt bridge between thesetwo residues in the prehydrolysis structure

MK MKT MK*DP MKD MK

K KT K*DP KD K

MK*T

K*T

MKDP

KDP

AMPPCP AMPPNP ADP-AlFx

ADP-Vi

ADP

<Strong-binding state> <Weak-binding state>

<Actively detaching state>

Fig. 1. Kinesin (KIF1A)–microtu-bule hydrolysis cycle. Abbrevia-tions: K, kinesin; M, microtubule;T, ATP; D, ADP; P, phosphate.Asterisks refer to the second ATPstate (K*T) or the second ADP-phosphate state (K*DP) of kine-sin. The main reaction pathwayis enclosed (45).

A

Neck-linker

Switch I

Switch II

α 4

β 9

L9L11

L12

B

α 5

Fig. 2. Crystal structures of KIF1A. (A) The AMP-PNP form of KIF1A. The switch I, switch II, andneck-linker regions are highlighted in red. (B) Superposition of AMP-PNP, ADP-AlFx, ADP-Vi, andADP structures. The AMP-PNP, ADP-AlFx, ADP-Vi, and ADP forms are shown in red, blue, green, andyellow, respectively.

R E P O R T S


closes this back door, which aligns the nucleo-philic water with the �-phosphate, and openingthis back door generates a tunnel through whichthe �-phosphate leaves the nucleotide-bindingpocket (20, 21). KIF1A resembles myosin inthe efficient release of the phosphate, which isgenerated by the swinging movement of theback door. The mutation of the back-door res-idue Arg216 significantly slows ATP hydroly-sis, which also supports the functional impor-tance of these back-door residues in fast ATPhydrolysis (22–25).

The conformational change of the back-door loop L9 not only contributes to the rapid

ATPase turnover, but also triggers large con-formational changes in the microtubule-binding loops in the switch II cluster. Beforehydrolysis, the conserved Gly251 at the N-terminal end of the switch II cluster forms ahydrogen bond with the �-phosphate (AMP-PNP form) (Fig. 3B) (fig. S2). The hydrolysisof ATP breaks this hydrogen bond and releas-es L11, the N-terminal loop of the switch IIcluster (ADP-AlFx form) (Fig. 3C). The re-lease of L11 triggers the rotation of the wholeswitch II cluster and the elongation of helix�4 from both ends by shortening and raisingof both loops L11 and L12 during the phos-

phate release (Fig. 3, C and D). The raisedconformation of loop L11 is stabilized by asalt bridge that forms between the conservedGlu267 in L11 and the conserved Arg216 inL9, and the hydrogen-bond network involv-ing the �-phosphate that connects switch Iloop L9 and switch II loop L11 (fig. S4) (26,27). Release of the phosphate breaks thesebonds, which results in partial melting of theraised loops L11 and L12 in the ADP form.The rotation of the switch II cluster is sup-ported by the salt bridge between the con-served Glu170 in loop L8 and the conservedArg316 in helix �5 (Fig. 3, E and F) (28). This

A

Neck-linkerSwitch I

Switch II

β 9

C

AMPPCP

G251

S215

ADP

AlF3

I354

L285

Switch IIα 4

α 6L9

Switch I

L11

Neck-linnker

4°

L9

Switch I

Switch IIα 4 L11

AMPPNP

G251

S215

Neck-linker

β 9I354

L285

α 6

ADP

α 4Switch II

Switch I

L11

R216

G262

E267Vi

L9

L285

I354α 6

G251

S215

B

D

Neck-linker

20°

L9

α 4L11

G251S215

I354

L285

α 6

E F

L8

Switch II

L11

L12

R316

E170

Fig. 3. The conformational chang-es that occur in the two switchregions and the neck-linker regionduring ATP hydrolysis. (A to D)AMP-PCP (A), AMP-PNP (B), ADP-AlFx (C), and ADP-Vi (D) forms areshown in light brown, red, blue,and green, respectively. Thesepanels are seen from the upperright in Fig. 2A (indicated by theblack arrow). Nucleotides and thecoordinating residues around themare shown as ball-and-stick mod-els. Missing C-terminal residuesand loops are shown by dashedlines. The structural details aroundthe nucleotide-binding pocket arealso shown in figs. S2 to S4 (46). (Eand F) The conserved salt bridgebetween Glu170 and Arg316 (en-closed by a red circle), shown inball-and-stick models in the AMP-PNP (E) and ADP-Vi (F) forms.Loop L8 and the switch II clusterare shown in dark blue and darkred, respectively. Nucleotides areshown as space-filling models.

R E P O R T S


rotational movement also leads to the dis-placement of the neck, as we have proposedpreviously (14, 29).

Thus, apparently, the rotational movementof the switch II cluster is supported by twoconserved salt bridges: that between Arg216

and Glu253 (back door), and that betweenGlu170 and Arg316. The back door mightserve as a “latch” to keep the switch II clusterto the strained “up” position. The release ofthe latch triggers the rotation of this helix,which is pulled to the relaxed “down” posi-tion by the salt bridge between Glu170 andArg316. The interaction between loop L8 andthe switch II cluster, mediated by the saltbridge between Glu170 and Arg316, mightserve as the mechanochemistry coupler thatcouples the ATPase chemical reaction to theconformational change of the microtubule-binding surface. Consistent with this hypoth-esis, mutation of Glu170 results in the decou-pling of the ATPase turnover from the de-tachment from the microtubule (6). Interest-ingly, a recent structural study with anotherkinesin KIF2C suggested that L8 might serveas the sensor for the microtubule, triggeringthe closure of the back door (30). The con-served salt bridge between L8 and �5 mightalso contribute to closing the back door, be-cause the mutation of either Glu170 or Arg316

considerably lowers the microtubule-activatedATPase activity (31).

Loop L11, helix �4, and loop L12 are themain microtubule-binding elements of kine-sin (4, 31–33); thus, their structural changesalter the affinity of kinesin to the microtu-bules. To examine the effect on microtubulebinding, we docked the atomic struct-ures of KIF1A and tubulin (34) to a KIF1A-microtubule complex whose structure was de-termined by cryogenic transmission electronmicroscopy (Fig. 4B) (14, 33). In the AMP-PNP form, loop L11 of KIF1A is extendeddown to the H11� helix of tubulin, and helix �4of KIF1A is inserted deep into the intradimergroove of tubulin. KIF1A is tilted toward L11so that L12 is extended up and away from themicrotubule. During the hydrolysis of ATP,both loops L11 and L12 are raised from themicrotubule, and only helix �4 supports theinteraction between KIF1A and the micro-tubule (ADP-Vi form). The rotation of the

Switch I

Switch IIL11 L12

R216

E253

A

S215

L11 L12

α 4

Microtubule

KIF1A

H11'-helix

B

E-hook

+end

α 4

K*-ATP(AMPPNP)

K-ADP-Pi(ADP-AlFx)

K*-ADP-Pi (ADP-Vi)

K-ADP (ADP)

Fig. 4. Hydrolysis-induced conformational changes on themicrotubule-binding surface of KIF1A. (A) Schematic illus-tration of the conformational changes in KIF1A. The switchI region (blue) and the switch II cluster (amino acids 251 to269 in red, 270 to 288 in yellow, and 289 to 305 in green)are shown as ribbon models. Nucleotides are shown asspace-filling models. The residues important for �-phosphate release (Ser215, Arg216, and Glu253) are shown asball-and-stick models. (B) The KIF1A-microtubule complexseen from the minus end of the microtubule. The micro-tubules (gray) are shown as space-filling models, except forhelices H11 and H12 and the E-hook (blue). See movies S1and S2 for details.

R E P O R T S


KIF1A core around helix �4 and the meltingof loop L12 into the flexible conformationthen allow loop L12 to interact with theC-terminal E-hook of tubulin in the ADPform while loop L11 is still raised.

These structural models suggest a thirdtransient mechanical state, in which kinesindetaches itself from the microtubule, betweenthe strong-binding and weak-binding states(Fig. 4B). In the ATP-bound state, kinesin istightly fixed to the microtubule by interac-tions between loop L11 of kinesin and theH11� helix of tubulin (the strong-bindingstate). ATP hydrolysis then generates the ac-tively detaching state by raising the two mainmicrotubule-binding loops L11 and L12.Thus, kinesin is only weakly supported byhelix �4 and other minor binding elements.After phosphate release, the flexible loop L12is extended down onto the tubulin C-terminalE-hook, and this flexible interaction allowsone-dimensional diffusion in the weak-binding state (4, 5, 32). In short, three suc-cessive states of the kinesin-microtubule in-teraction exist during ATP hydrolysis, andkinesin uses the energy of ATP hydrolysis todetach loop L11 from the microtubule and toextend another loop (L12) to the microtubule.

We tested this proposal of alternating use ofthe two microtubule-binding loops by introduc-ing mutations to these loops. The positivelycharged residues Lys261, Arg264, and Lys266 inloop L11 were mutated to Ala, and this L11mutant (K261A/R264A/K266A) was comparedwith an L12 mutant (CK1) (4) and an L8 mu-tant in which Lys161, Arg167, Arg169, andLys183 were mutated to Ala (K161A/R167A/R169A/K183A). Loop L8 was selected for mu-tation because it corresponds to the third micro-tubule-binding region and does not change itsconformation or position relative to the micro-tubule during ATPase turnover (14, 33). Rela-tive to the wild-type protein, all of the mutantsshowed similar ATPase turnover rates but alower affinity to the microtubule. As previouslyshown (4), the mutation in loop L12 selectivelyweakened affinity in the ADP state but not inthe AMP-PNP state ( Table 1). By contrast, themutation in loop L11 had a stronger effect inthe AMP-PNP state than in the ADP state. Themutation in loop L8 affected both forms to thesame extent. Thus, the mutational studies sup-port the proposal that KIF1A uses two loops,L11 and L12, in an alternating manner duringATPase turnover.

This alternating use of microtubule-binding loops provides structural support forour model of KIF1A’s motility (4, 5, 32, 35).The model assumes that KIF1A alternatesbetween two different binding states: the rig-or-binding state with the affinity biased to-ward the forward tubulin subunit, and thediffusive binding state that allows one-dimensional diffusion. The rigor-bindingloop L11 is extended toward the forward

tubulin subunit. This configuration of loopL11 gives directional bias, or asymmetricbinding potential, to the rigor-binding state.The force and binding energy supported bythis binding are more than 10 pN and 10 kBT(3, 4), respectively. This strong binding isbroken by the energy of ATP hydrolysis.ATP hydrolysis causes conformationalchanges that first raise loop L11 from themicrotubule and then bring loop L12 intocontact with the microtubule. The flexibletether between loop L12 and the E-hook oftubulin allows diffusive binding in the ADPform. Thus, KIF1A kinesin takes three struc-tures successively: L11 down, both L11 andL12 up, and L12 down. It is therefore possi-ble to surmise that a fourth structure mightexist: both L11 and L12 down (at the transi-tion from the L12 down, ADP state, to theL11 down, ATP state). Bringing both L11and L12 to the microtubule might open thenucleotide-binding pocket and accelerateADP release. Further structural analysis ofthe nucleotide-free state or the ADP-releaseintermediate state of KIF1A will be requiredto examine this possibility.

Finally, we note that the design principleof KIF1A is similar to that of heterotrimericguanine nucleotide–binding proteins (G pro-teins) (36, 37) and other proteins such asprotein kinases (38). The binding energy ofKIF1A to the microtubule is used to achievea mechanical step of KIF1A (5). With Gproteins and protein kinases, the binding en-ergy to an effector molecule or substrate isused to cause structural changes. Nucleophil-ic attack by water molecules or serine resi-dues on the �-phosphate catalyzes hydrolysis.Hydrolysis triggers a conformational changeat the binding surface (the switch II helix)that detaches the enzyme from the targetmolecule. The binding between a G proteinand its effector molecule, or between kinaseand its substrate, is so strong that energy isrequired for dissociation. The energy ofhydrolysis is used for this active detach-ment, which is necessary to cycle the reac-tion. This conserved strategy might reflectthe evolutionary pathway of these classesof proteins, and also suggests a design prin-ciple for nanomachines.

References and Notes1. N. Hirokawa, Science 279, 519 (1998).2. M. Nishiyama, E. Muto, Y. Inoue, T. Yanagida, H.

Higuchi, Nature Cell Biol. 3, 425 (2001).3. M. J. Schnitzer, K. Visscher, S. M. Block, Nature Cell

Biol. 2, 718 (2000).4. Y. Okada, N. Hirokawa, Proc. Natl. Acad. Sci. U.S.A.

97, 640 (2000).5. Y. Okada, H. Higuchi, N. Hirokawa, Nature 424, 574

(2003).6. L. M. Klumpp, A. Hoenger, S. P. Gilbert, Proc. Natl.

Acad. Sci. U.S.A. 101, 3444 (2004).7. I. M. Crevel et al., EMBO J. 23, 23 (2004).8. Although the biochemically measured apparent dis-

sociation constant is not significantly differentamong the AMP-PNP–bound, nucleotide-free, and

ADP-bound states, each kinesin molecule behavesdifferently on the microtubule in the ADP-boundstate (one-dimensional diffusion) relative to the oth-er two states (rigor-binding) (4, 5). Biochemical anal-ysis using [�-32P]ADP indicates that more than 90%of the active sites of kinesin molecules are occupiedwith ADP. These results, as well as previous resultswith the L12 mutant CK1 (4), support that the resultwith excess ADP (2 to 10 mM) reflects the behaviorof kinesin with ADP in the active site.

9. Our previous experiment with ADP plus phosphatefailed to detect the actively detaching state (4). Thismight be due to the irreversibility of the phosphaterelease. As reported with myosin (39), our prelimi-nary biochemical results and our current structuralstudy (fig. S4) suggest that exogenously added phos-phate apparently does not bind to KIF1A. Thus, in ourprevious experiments with ADP and phosphate, mostof the KIF1A will be in the ADP-bound state, not inthe ADP-phosphate state. That is why we used ana-logs, which are expected to better mimic the trueintermediate states of ATPase turnover.

10. The increase in the value of the dissociation constant(Kd) of the ADP-Vi state relative to other statessuggests that the dissociation of kinesin is accelerat-ed by a factor of at least 5. Our preliminary resultswith single-molecule imaging suggest that the off-rate is faster in the ADP-Vi state than in other statesby a factor of more than 300.

11. E. Pechatnikova, E. W. Taylor, Biophys. J. 77, 1003(1999).

12. H. J. Matthies, R. J. Baskin, R. S. Hawley,Mol. Biol. Cell12, 4000 (2001).

13. S. S. Rosenfeld, B. Rener, J. J. Correia, M. S. Mayo, H. C.Cheung, J. Biol. Chem. 271, 9473 (1996).

14. M. Kikkawa et al., Nature 411, 439 (2001).15. J. F. Koretz, E. W. Taylor, J. Biol. Chem. 250, 6344 (1975).16. The RMSDs of the switch I region (amino acids 202 to

218) are 2.0 Å (AMP-PNP versus ADP-AlFx forms), 4.5Å (ADP-AlFx versus ADP-Vi forms), and 4.0 Å (ADP-Viversus AMP-PNP forms). Those of the switch II region(amino acids 248 to 324) are 0.8 Å (AMP-PNP versusADP-AlFx forms), 1.4 Å (ADP-AlFx versus ADP-Viforms), and 2.1 Å (ADP-Vi versus AMP-PNP forms).

17. Although the �-phosphate is expected to be nucleo-philically attacked by water, by analogy to myosin(20, 40), the �-phosphate in our transition structurewith ADP-AlFx coordinates with Ser215 in the switchI region. This coordination could possibly be an arti-fact due to the use of a nucleotide analog. However,the difference map of AlF3 indicates that Al

3� existsin trigonal bipyramidal configuration with partialbonds to the �-phosphate and the hydroxyl group ofSer215 (Fig. 3C) (fig. S3). This coordination suggestsanother possibility: that the �-phosphate is nucleo-philically attacked by conserved Ser215, as occurs inprotein kinases. Further studies will be necessary toclarify this issue.

18. The conserved residue Arg216 in the switch I loop L9has been suggested to be the important phosphate-binding site near the nucleotide-binding pocket (41).

19. The possibility that our structure of ADP-Vi formmight only reflect a structural variation of ADP isexcluded by the following results. First, ADP-boundKIF1A never takes an ADP-Vi like structure, evenunder conditions identical or similar to the crystalli-zation condition for the ADP-Vi–bound KIF1A. Simi-larly, ADP-Vi–bound KIF1A never takes an ADP-likestructure, even under the crystallization condition forthe ADP form. This is very different from the casewith the ADP and AMP-PCP forms (14). ADP-boundKIF1A can take an AMP-PCP–like structure (but notan AMP-PNP–like structure) under the crystallizationcondition for the AMP-PCP form. This reflects thatthe energy barrier for the transition between theADP-Vi and ADP structures is much higher than thatbetween the ADP and AMP-PCP structures. Second,only preformed ADP-Vi-KIF1A complex takes thisADP-Vi structure. We could not obtain a crystal ofADP-Vi form by soaking of vanadate solution toADP-KIF1A crystals or by the addition of vanadate tothe mother liquid with which ADP-bound KIF1A iscrystallized. These results strongly suggest that ourADP-Vi structure reflects a structure of a stable in-termediate state within the ATPase cycle, rather than

R E P O R T S


another structural variant or an artificial vanadateadduct to ADP-KIF1A.

20. R. G. Yount, D. Lawson, I. Rayment, Biophys. J. 68,47S (1995).

21. M. Furch, S. F. Becker, M. A. Geeves, K. C. Holmes, D. J.Manstein, J. Mol. Biol. 290, 797 (1999).

22. M. Yun, X. Zhang, C. G. Park, H. W. Park, S. A. Endow,EMBO J. 20, 2611 (2001).

23. C. M. Farrell, A. T. Mackey, L. M. Klumpp, S. P. Gilbert,J. Biol. Chem. 277, 17079 (2002).

24. Recently, a new model was proposed for myosin onthe basis of a new structure (42). Our structures andmodel are not inconsistent with this new mechanism.As proposed for the new “trapdoor” mechanism,kinesin has two “doors” that open and close asyn-chronously. The door on the switch I side is openedand the door on the switch II side is closed (“O/Cconformation”) in the preisomerization state. Thisconformation might contribute to the holding of theentering ATP. Before hydrolysis, KIF1A takes a “C/Cconformation.” During hydrolysis, it transiently takesa “C/O conformation” for the effective nucleophilicattack. Then, it takes an “O/O conformation,” whichopens the route through which the phosphate isreleased. Further studies will be necessary to deter-mine whether kinesin works by a back-door or atrapdoor mechanism.

25. Here, two differences between kinesins and myosinsshould be noted: (i) The nucleotide-binding pocket ofmyosin is enclosed, whereas that of kinesin is ratherexposed to the surface; (ii) the back door in theprehydrolysis state of myosin is rigidly closed, where-as that of kinesin is only partially closed. Thesedifferences might affect the kinetics of kinesin AT-Pase much faster than those of myosins (32, 43).Further structural study of the KIF1A-microtubulecomplex will be needed to clarify these differences.

26. The possibility that the vanadate moiety on Arg216

only reflects an artificial coordination of the solvention in crystal can be excluded by the following rea-sons. First, the position of vanadate is analogous tothat in myosin. With myosin, vanadate is found onthe corresponding back-door residue through whichthe produced phosphate is thought to be released.Second, vanadate is located at the same or similarposition even in solution. The position of vanadate insolution was determined by the photo cleavage ex-periment. With myosins, the cleavage site (P-loop)agrees well with its ADP-Vi structure (40). Similarly,KIF1A was cleaved at loop L11, as expected from ourADP-Vi structure. This confirms that our ADP-Vistructure actually reflects the solution structure.

27. This stabilization of loop L11 of the ADP-Vi form isalso evident from the low temperature factors of theresidues in loop L11 [mean B value of the main-chainatoms of loop L11 (amino acids 251 to 270), 39.5 Å2,overall mean B value, 32.0 Å2]. In the ADP form, theresidues from Ala255 to Ala260 in loop L11 are disor-dered because of its flexibility.

28. This arginine is conserved among kinesins, althoughsome kinesins have arginine at the residue corre-sponding to 313, a position just next to 316 on helix�5. For example, in conventional kinesin (3KIN),Arg313 forms a salt bridge to Glu170.

29. Before ATP hydrolysis, the neck-linker �9 (aminoacids 353 to 361) is fully docked [except for the(His)6 tag at the C terminus] to the catalytic core(AMP-PCP and AMP-PNP forms) (Fig. 3, A and B).During hydrolysis, the rotational movement of theswitch II helix �4 pushes the neck-linker �9 downthrough the hydrophobic interaction between theconserved Leu285 of helix �4 and the conserved Ile354

of strand �9 (ADP-AlFx form) (Fig. 3C). This leads tothe complete detachment of the neck-linker from thecatalytic core (ADP-Vi and ADP forms) (Fig. 3D).

30. T. Ogawa, R. Nitta, Y. Okada, N. Hirokawa, Cell 116,485 (2004).

31. G. Woehlke et al., Cell 90, 207 (1997).32. Y. Okada, N. Hirokawa, Science 283, 1152 (1999).33. M. Kikkawa, Y. Okada, N. Hirokawa, Cell 100, 241

(2000).34. J. Lowe, H. Li, K. H. Downing, E. Nogales, J. Mol. Biol.

313, 1045 (2001).35. This active detachment mechanism is also applica-

ble to dimeric kinesin, including conventional kine-

sin. Hydrolysis of one head triggers the conforma-tional change of its switch II cluster, both themicrotubule-binding loop raised and the switch IIhelix rotated. The rotation of this helix pushes theneck-linker by which two heads coordinate. Thus,our study provides structural evidence of the ideathat the active detachment is coupled with thecoordination of two heads (44). In other words,active detachment of one head precedes the hy-drolysis of the second head, the second step.

36. S. R. Sprang, Annu. Rev. Biochem. 66, 639 (1997).37. J. J. Tesmer, S. R. Sprang, Curr. Opin. Struct. Biol. 8,

713 (1998).38. D. A. Johnson, P. Akamine, E. Radzio-Andzelm, M.

Madhusudan, S. S. Taylor, Chem. Rev. 101, 2243 (2001).39. C. F. Midelfort, Proc. Natl. Acad. Sci. U.S.A. 78, 2067

(1981).40. C. A. Smith, I. Rayment, Biochemistry 35, 5404 (1996).41. C. V. Sindelar et al., Nature Struct. Biol. 9, 844 (2002).42. T. F. Reubold, S. Eschenburg, A. Becker, F. J. Kull, D. J.

Manstein, Nature Struct. Biol. 10, 826 (2003).43. M. A. Geeves, Biochem. J. 274, 1 (1991).44. A. T. Mackey, S. P. Gilbert, Biochemistry 39, 1346 (2000).45. Both conventional kinesin and KIF1A share this path-

way. The only major difference is the affinity ofkinesin to microtubule in the ADP state (yellow). TheK-loop of KIF1A increases its affinity in the ADP state(4), shifting the equilibrium to the associated state

(MKD). With K-loop mutant KIF1A as well as conven-tional kinesin, the equilibrium is shifted to the disso-ciated state (KD).

46. See supporting material on Science Online.47. We thank D. R. Tomchick, N. Ota, S. Wakatsuki, N.

Igarashi, H. Sawa, Y. Wakabayashi, and their col-leagues for technical suggestions; H. Fukuda, H. Sato,and M. Sugaya for technical and secretarial assist-ance; and H. Miki, M. Kawagishi, H. Yajima, T. Ogawa,E. Nitta, and our colleagues for discussions and tech-nical assistance. Supported by a Center of ExcellenceGrant-in-Aid from the Ministry of Education, Culture,Sports, Science and Technology of Japan (N.H.). Co-ordinates for the four new structures of KIF1A havebeen deposited in the PDB database and are availableunder the following accession codes: both 1VFV and1VFW for the AMP-PNP form, 1VFX for the ADP-AlFxform, and 1VFZ for the ADP-Vi form.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/305/5684/678/DC1Materials and MethodsTable S1Figs. S1 to S4Movies S1 and S2References

10 February 2004; accepted 16 June 2004

Crystal Structures of HumanCytochrome P450 3A4 Bound toMetyrapone and ProgesteronePamela A. Williams,1 Jose Cosme,1 Dijana Matak Vinkovic,1

Alison Ward,1* Hayley C. Angove,1 Philip J. Day,1

Clemens Vonrhein,2 Ian J. Tickle,1 Harren Jhoti1†

Cytochromes P450 (P450s) metabolize a wide range of endogenous com-pounds and xenobiotics, such as pollutants, environmental compounds, anddrug molecules. The microsomal, membrane-associated, P450 isoformsCYP3A4, CYP2D6, CYP2C9, CYP2C19, CYP2E1, and CYP1A2 are responsiblefor the oxidative metabolism of more than 90% of marketed drugs. Cyto-chrome P450 3A4 (CYP3A4) metabolizes more drug molecules than all otherisoforms combined. Here we report three crystal structures of CYP3A4:unliganded, bound to the inhibitor metyrapone, and bound to the substrateprogesterone. The structures revealed a surprisingly small active site, withlittle conformational change associated with the binding of either com-pound. An unexpected peripheral binding site is identified, located above aphenylalanine cluster, which may be involved in the initial recognition ofsubstrates or allosteric effectors.

CYP3A4 is a complex heme-containing en-zyme that exhibits non–Michaelis-Menten ki-netics and both homotrophic and heterotro-phic cooperativity toward several substrates(1, 2). For example, CYP3A4 exhibits atyp-ical kinetic behavior in vitro with aflatoxinB1, amitriptyline, cabamazepine, proges-

terone, and diazepam (3, 4). Studies havesuggested that CYP3A4 has a noncatalyticeffector site within the active-site cavity, ca-pable of modulating function. In this model,the substrate and effector molecules occupyseparate positions and only the substrate mol-ecule has access to the reactive oxygen (5).These observations, combined with the vastchemical diversity of CYP3A4 substrates,have led to speculation that CYP3A4 has anadaptive active site capable of binding mul-tiple compounds simultaneously (6–10). Af-ter examining the interaction of 10 probesubstrates and finding that they could be di-vided into groups related to their size orchemical class, Kenworthy et al. (11) pro-

1Astex Technology, 436 Cambridge Science Park, Mil-ton Road, Cambridge, CB4 0QA, UK. 2Global PhasingLtd., Sheraton House, Castle Park, Cambridge, CB30AX, UK.

*Present address: AstraZeneca, Research and Devel-opment, Charnwood, Bakewell Road, Loughborough,LE11 5RH, UK.†To whom correspondence should be addressed. E-mail: [email protected]

R E P O R T S


posed that three subpockets exist within theactive site of CYP3A4. Additional studieshave demonstrated the existence of multiplekinetically distinguishable conformations ofCYP3A4, in the presence or absence of asubstrate, effector, or inhibitor molecule(s)(12–14), which may contribute to allostery.External factors such as the binding of cyto-chrome b5 and the presence of phospholipids(15) also affect the activity of CYP3A4 to-ward some compounds and may also stabilizeparticular conformations.

Understanding how this complex enzymerecognizes multiple, diverse chemical struc-tures has been hampered by the lack of knowl-edge about the three-dimensional structure ofCYP3A4. Homology models (16, 17) have lim-ited use because of the substantial sequencediversity of P450s. The most widely used tem-plates for modeling CYP3A4 are the P450structures from Bacillus megaterium (P450BM3) and Saccharopolyspora erythraea (P450EryF), both of which share less than 25% se-quence identity with CYP3A4. Furthermore,the sequence identity with human CYP2C9, forwhich the crystal structure was recently deter-mined (18), is only 24%. We report crystalstructures of CYP3A4, both unliganded andbound to either the inhibitor metyrapone or thesubstrate progesterone. The protein used inthese studies was N-terminally truncated to fa-

cilitate crystallization and shown to be function-ally active (19).

The ligand-free structure of CYP3A4was solved to a resolution of 2.8 Å by acombination of multiple-wavelength anom-alous dispersion and molecular replacementtechniques (19) (tables S2 and S3). Theoverall structure of CYP3A4 conforms tothe fold that is characteristic of the P450superfamily, with a small, predominantly�-strand N-terminal domain and a larger,helical, C-terminal domain that contains theheme and the active site (Fig. 1A). Theheme iron is ligated by a conserved cys-teine (Cys442) and the propionates of theheme interact with the side chains ofArg105, Trp126, Arg130, Arg375, and Arg440.The heme is accessible to solvent by meansof channels formed by � sheet 1, the B-Cloop and the F-G region (Fig. 1B and fig.S1). There is a channel running from theactive site to the outside of the proteinformed by the B-C loop and the G� helix,whereas another channel is formed by theF� helix and the � sheet 1. The two chan-nels are separated by the loop between theB� and the C helices, a region well recog-nized to adopt different conformations de-pendent on the presence of a ligand (20).Although the heme of the protein sampleused for crystallization was ferric and low spin,

x-rays are known to be capable of reducing hemeiron, and thus its oxidation state in the structureis unknown (21). A low-spin ferric heme ironwould be expected to have a ligand or watermolecule in the sixth coordination position, butthe structure reveals no ordered water bound inthis position. A five-coordinate heme iron wouldrequire that the iron be displaced out of the planeof the coordinating porphyrin nitrogen atomsand toward the cysteine sulfur atom. However,given the resolution of the data, no conclusionscan be reliably drawn regarding coplanarity ofthe heme iron. Another functionally importantwater molecule in other P450s is located be-tween the highly conserved residues Ala305 andThr309, which lie on the I helix and are close tothe heme. In some P450 structures, the threonineresidue interacts with this water molecule, whichdisrupts the hydrogen-bonding pattern of the sec-ondary structure, resulting in a kink in helix I(22). Although a similar but less pronouncedkink is observed in CYP3A4, there is no discretedensity visible for this water molecule.

CYP3A4 has a number of features thatdiffer from other P450 structures. A nota-bly hydrophobic region is located aroundthe loop following helix A�� (top left, Fig.1A), a helix not observed in other P450crystal structures. This region, along withthe G� helix and the loop between the G�and G helices, which are also hydrophobic,

Fig. 1. (A) Overall fold ofCYP3A4, colored fromblue at the N terminus togreen, to yellow, to redat the C terminus. Thesame coloring scheme isused for CYP3A4 in (B)and in Figs. 3 and 4 andfor CYP2C9 in (C). Theheme is depicted as aball-and-stick model inthe center of the mole-cule, flanked by helix I.Helix A��, not present inother P450 structures, isshown at the top left ofthe structure, and theshortened F helix is tothe right of the heme. Allfigures were producedwith AstexViewer version2 (36). (B) View of thePhe-cluster of CYP3A4.The solvent-accessiblemolecular surface of the active-surface cavity of CYP3A4 is depicted as asemitransparent green surface. The seven phenylalanine residues that makeup the Phe-cluster are depicted as ball-and-stick models. The availablevolume of the active site is about 520 Å3 [calculated with the LIGSITEalgorithm (28)]. The short F helix is shown at the right of the structure, andleads into a loop region that contains Phe213, Phe215, Phe219, and Phe220

before the F� helix. The Phe-cluster is about 10 Å above the heme, which isdepicted as a ball-and-stick model at the bottom of the structure. (C) Viewof the active-site cavity of CYP2C9. The solvent-accessible molecular surfaceof CYP2C9 (PDB entry 1OG2) is depicted as a semitransparent surface inpurple. The available volume of the active site is about 600 Å3 [calculatedwith the LIGSITE algorithm (28)]. The F helix of CYP2C9 is longer than that ofCYP3A4, and the relative position of secondary structural elements results ina narrowing of the active site toward the heme and a substantially different shape when compared with the active site of CYP3A4.

R E P O R T S


may mediate interaction with the microso-mal membrane. Another unexpected featureof the CYP3A4 structure is the region fol-lowing helix F (Fig. 1, A and B). This helixis markedly short compared with otherP450 structures, and it leads into an orderedstretch of polypeptide chain that does notconform to any secondary structure motif.This region is located above and perpendic-ular to helix I and contains a number ofresidues that have been shown by site-directed mutagenesis to have a direct orindirect role in CYP3A4 function. For ex-ample, Leu210, implicated in effector bind-ing as well as the stereo- and regioselectiv-ity profile of CYP3A4 (7, 23, 24 ), lies inthis region with its side chain pointingaway from the active site. Leu211 andAsp214, implicated in cooperativity (23, 25),also lie in this extended loop region and pointaway from the active site. In contrast, the sidechains of two other residues shown by mu-tagenesis to have no role in cooperativity (23),Phe213 and Phe215, point toward the active siteand, together with another five phenylalanineresidues (Phe108, Phe219, Phe220, Phe241, andPhe304), form a “Phe-cluster” (Fig. 1B). No

other structurally characterized P450 has such acluster of phenylalanine residues in this region.Some of these other Phe-cluster residues havealso been shown by mutagenesis studies to beinvolved in CYP3A4 activity. For example,Phe304, which lies on the I helix, has a dualrole in cooperativity, regioselectivity, and ste-reoselectivity (7, 26, 27), whereas the substitu-tion of Phe108 with a smaller or larger ami-no acid affected the metabolism of somesubstrates (7).

The Phe-cluster lies above the activesite, with the aromatic side chains stackingagainst each other to form a prominenthydrophobic core. This region of the struc-ture is highly ordered with an average Bfactor of 41 Å2 for the residues in thecluster, compared with the average of 66Å2 over the entire structure. Furthermore,as a result of this aromatic clustering, theactive site of CYP3A4 has a probe-accessible volume (28) of about 520 Å3,substantially smaller than expected on thebasis of its reported metabolism of largesubstrates. The overall volume of theCYP3A4 active site is also smaller thanthat of CYP2C9 (18), which is 670 Å3, and

different in shape (Fig. 1, B and C). Thevariation in the active-site topologies is aconsequence of the Phe-cluster, which ispositioned on the top of the active sitecloser to the heme, with � sheet 1 fartherfrom the heme compared with CYP2C9(Fig. 1C). An overlay of CYP3A4 with thestructures of P450 BM3 (29) and P450EryF (30) (Fig. 2) shows the limitations ofhomology models based on these two bac-terial structures. The position of the Phe-cluster results in a closed conformation ofthe active site of CYP3A4 that is similar tothe closed conformation adopted by P450EryF (30), but the spatial arrangement ofthe F-G loop and F helices is not conserved(Fig. 2). The heme of CYP3A4 has greateraccessibility to the active site than does thatof CYP2C9 (Fig. 1C), which could allowtwo substrate molecules to have access tothe reactive oxygen, consistent with dataindicating that CYP3A4 is able to bind andmetabolize multiple substrate molecules si-multaneously (6, 8). Conformational move-ment involving the Phe-cluster could alsoreposition phenylalanine residues, perhapsresulting in an extension of helix F and alarger active site. There is a precedent forsuch a mechanism from the CYP119 struc-ture in which, although there is no analo-gous Phe-cluster, ligand binding resulted inan extension of helix F and movement ofthe G helix and the intervening F-G loop(31). To investigate whether conformation-al movement is a necessary prerequisite forligand binding by CYP3A4, we determinedthe crystal structure in complex with theknown inhibitor metyrapone. The metyrap-one cocomplex structure was determinedwith the use of both soaking and, moreimportantly, cocrystallization techniques,which allowed conformational movementby the protein if required.

Contrary to expectations, the binding ofmetyrapone to CYP3A4 reveals essentiallyno conformational changes in the protein.Ultraviolet and visible spectroscopic data(32) indicate that in solution the inhibitorligates the heme. This is consistent with thebinding mode observed in the crystal struc-ture, in which metyrapone is bound directlyto the heme iron by means of the alkyl-pyridine nitrogen (Fig. 3), and exhibits goodshape complementarity with the CYP3A4 ac-tive site (fig. S2). An alternative bindingmode for metyrapone may involve the nitro-gen atom of the keto-pyridine group, as ob-served previously in the cocomplex with bac-terial P450 cam (33). However, the electrondensity for this CYP3A4 cocomplex indicatesthat the illustrated binding mode is predomi-nant. The volume occupied by the metyrap-one molecule is 230 Å3, leaving sufficientspace for additional molecules to bind withinthe active site.

Fig. 2. Stereoimage showing the structural overlay of CYP3A4 in red, with P450 BM3 (molecule Aof PDB entry 1BU7) in blue, and P450 EryF (PDB entry 1OXA) in yellow. The heme is depicted asa ball-and-stick model toward the bottom of the figure. The F helix of CYP3A4 (red) is notablyshorter than the F helices of P450 BM3 (blue) or P450 EryF (yellow). CYP3A4 has two helices F� andG� that are absent in P450 BM3, P450 EryF, and other bacterial P450 structures.

Fig. 3. Stereoview of metyrapone bound to the heme of CYP3A4. The metyrapone, heme, andsurrounding residues are depicted as ball-and-stick models. The initial sigmaA-weighted Fo-Fcelectron density map is shown in red contoured at the 2.0� level, and the final sigma-weightedFo-Fc electron density map, calculated omitting the heme, is shown in red contoured at the2.0� level, and then is shown in blue contoured at the 1.0� level. The compound exhibits asingle binding orientation and interacts with the protein by means of a pyridinyl groupof metyrapone.

R E P O R T S


Although formation of the metyraponecomplex is achieved with little conforma-tional change, we anticipate that substantialprotein movement, possibly involving the Fand G helical regions and the Phe-cluster,would be required to permit the entry andaccommodation of larger compounds.Movement of the B-C loop region couldalso open up the channel and increase thevolume of the active site. To investigatethis possibility and explore how CYP3A4recognizes substrates, we determined thestructure of the complex with progesteroneby cocrystallization. Surprisingly, proges-terone also induced very little conforma-tional change in the protein and bound at aperipheral site very close to the Phe-cluster(Fig. 4) and some distance away (�17 Å)from the heme iron. The initial electrondensity maps indicated that one progester-one molecule bound in a single clearlydefined orientation (fig. S3), as reflected byan average B factor for progesterone of 45Å2, compared with 61 Å2 for the protein.The progesterone molecule interacts withthe protein through a hydrogen bond be-tween its acetyl oxygen and the amide ni-trogen of Asp214, packs against the sidechains of Phe219 and Phe220 in the Phe-cluster (fig. S3), and appears fairly acces-sible to solvent. Although this binding sitemay be an artifact of crystallization, weargue that it is likely to have functionalrelevance. This is supported by the obser-vation that the side chains of Leu211,Phe213, and Asp214 residues implicated inboth progesterone homotrophic and hetero-trophic cooperativity (5, 25) lie in the vi-cinity of this peripheral binding pocket. Onthe basis of these findings, we propose thatthe progesterone-binding site may beinvolved in the recognition of effector aswell as substrate molecules and has a rolein modulating cooperativity. Notably, al-

though progesterone can be docked into theactive site of CYP3A4 in an orientationconsistent with its metabolism, there is noevidence of progesterone binding withinthe active-site cavity in this structure.

Several studies indicate that the F-G andB-C loops of P450, which constitute theputative substrate-access channel, couldcontribute to membrane binding and orientthe substrate-access channel toward themembrane surface (34, 35 ). The location ofthe progesterone-binding site in the F-Gregion is consistent with a role in initialsubstrate recognition, because it lies alongan appropriate route for compounds thatmove directly from the membrane into theactive site of CYP3A4. Appropriate confor-mational movement of residues around thePhe-cluster could result in a substrate-access channel that stretched from this pe-ripheral binding site to the heme group,providing a route for a compound to movefrom this initial recognition site to the ac-tive site. Such conformational movementsin the Phe-cluster could be triggered byinteraction with a physiological electrontransfer partner such as cytochrome b5 orcytochrome P450 reductase, or by a changein membrane properties to allow the sub-strate molecule to move from its surfaceposition to a position in the active site thatis suitable for metabolism. Future studiesthat investigate these potential mechanismsof molecular recognition and enzyme activ-ity can now be guided by the crystal struc-ture of CYP3A4.

References and Notes1. F. P. Guengerich, Annu. Rev. Pharmacol. Toxicol. 39, 1(1999).

2. J. M. Hutzler, T. S. Tracy, Drug Metab. Dispos. 30, 355(2002).

3. Y. F. Ueng, T. Kuwabara, Y. J. Chun, F. P. Guengerich,Biochemistry 36, 370 (1997).

4. G. E. Schwab, J. L. Raucy, E. F. Johnson, Mol. Pharma-col. 33, 493 (1988).

5. T. L. Domanski et al., Biochemistry 40, 10150 (2001).6. K. R. Korzekwa et al., Biochemistry 37, 4137 (1998).7. K. K. Khan, Y. Q. He, T. L. Domanski, J. R. Halpert, Mol.Pharmacol. 61, 495 (2002).

8. M. Shou et al., Biochemistry 33, 6450 (1994).9. K. E. Kenworthy, S. E. Clarke, J. Andrews, J. B. Houston,Drug Metab. Dispos. 29, 1644 (2001).

10. P. Lu et al., Drug Metab. Dispos. 29, 1473 (2001).11. K. E. Kenworthy, J. C. Bloomer, S. E. Clarke, J. B.

Houston, Br. J. Clin. Pharmacol. 48, 716 (1999).12. A. P. Koley, J. T. Buters, R. C. Robinson, A. Markowitz,

F. K. Friedman, J. Biol. Chem. 270, 5014 (1995).13. A. P. Koley, J. T. Buters, R. C. Robinson, A. Markowitz,

F. K. Friedman, J. Biol. Chem. 272, 3149 (1997).14. D. R. Davydov, J. R. Halpert, J. P. Renaud, H. G. Hui

Bon, Biochem. Biophys. Res. Commun. 312, 121(2003).

15. K. H. Kim, T. Ahn, C. H. Yun, Biochemistry 42, 15377(2003).

16. D. F. Lewis, P. J. Eddershaw, P. S. Goldfarb, M. H.Tarbit, Xenobiotica 26, 1067 (1996).

17. G. D. Szklarz, J. R. Halpert, J. Comput. Aided Mol. Des.11, 265 (1997).

18. P. A. Williams et al., Nature 424, 464 (2003).19. Materials and methods are available as supporting

material on Science Online.20. T. L. Poulos, Proc. Natl. Acad. Sci. U.S.A. 100, 13121

(2003).21. C. Nave, Radiat. Phys. Chem. 45, 483 (1995).22. D. C. Haines, D. R. Tomchick, M. Machius, J. A. Peter-

son, Biochemistry 40, 13456 (2001).23. G. R. Harlow, J. R. Halpert, J. Biol. Chem. 272, 5396

(1997).24. H. Wang et al., Biochemistry 37, 12536 (1998).25. G. R. Harlow, J. R. Halpert, Proc. Natl. Acad. Sci.

U.S.A. 95, 6636 (1998).26. T. L. Domanski, J. Liu, G. R. Harlow, J. R. Halpert, Arch.

Biochem. Biophys. 350, 223 (1998).27. J. C. Stevens et al., J. Pharmacol. Exp. Ther. 290, 594

(1999).28. M. Hendlich, F. Rippmann, G. Barnickel, J. Mol. Graph.

Model. 15, 355 (1997).29. I. F. Sevrioukova, H. Li, H. Zhang, J. A. Peterson, T. L.

Poulos, Proc. Natl. Acad. Sci. U.S.A 96, 1863 (1999).30. J. R. Cupp-Vickery, T. L. Poulos, Nature Struct. Biol. 2,

144 (1995).31. S. Y. Park et al., J. Inorg. Biochem. 91, 491 (2002).32. P. A. Williams et al., data not shown.33. T. L. Poulos, A. J. Howard, Biochemistry 26, 8165

(1987).34. J. Cosme, E. F. Johnson, J. Biol. Chem. 275, 2545

(2000).35. S. E. Graham-Lorence, J. A. Peterson, in Physiological

Functions of Cytochrome P450 in Relation to Struc-ture and Regulation, C. Jefcoate, Ed. (Jai Press, Green-which, CT, 1995), pp. 57–79.

36. M. J. Hartshorn, J. Comput. Aided Mol. Des. 16, 871(2002).

37. We thank S. Rich, S. Maman, D. Cross, S. Kirton, andC. Murray for assistance with crystallization, charac-terization, and analysis; G. Williams for discussions;and F. W. Dahlquist for the provision of the pCWori�expression vector. We acknowledge the EuropeanSynchrotron Radiation Facility for the provision ofsynchrotron radiation facilities and thank the indus-trial user group for assistance with data collection.Coordinates and structure factors of the ligand-free,metyrapone-, and progesterone-bound structures ofCYP3A4 have been deposited in the Protein DataBank (PDB) (entries 1W0E, 1W0F, and 1W0G).

Supporting Online Materialwww.sciencemag.org/cgi/content/full/1099736/DC1Materials and MethodsFigs. S1 to S3Tables S1 to S3References

29 April 2004; accepted 7 July 2004Published online 15 July 2004;10.1126/science.1099736Include this information when citing this paper.

Fig. 4. Peripheral bind-ing of progesterone toCYP3A4. The solvent-accessible surface of theprogesterone moleculeis shown in yellow at thetop of the figure. A crosssection of the solvent-accessible surface ofCYP3A4 is shown inblue, revealing theactive-site cavity ofCYP3A4 in the center ofthe figure. Progesteronesits in a shallow pocketon top of the Phe-clusterof CYP3A4, separatedfrom the heme by theF-G region of the mole-cule. Progesterone binds17 Å above the hemeiron and on the face of CYP3A4 most likely to interact with the membrane.

R E P O R T S



ELSD HPLC IMPROVEMENTNew remote control software forthe Chromachem Evaporative LightScattering Detector (ELSD) simpli-fies detector operation and high-performance liquid chromatography

(HPLC) system integration.The Chromachem ELSD provides univer-sal detection of non-volatile compounds. Unlike HPLC ultraviolet orfluorescence detectors, the Chromachem ELSD does not require thepresence or absence of chromophores, fluorophores, or electroac-tive groups for detection. The technique is fully compatible withboth organic solvent–based and aqueous mobile phases, even athigh or microbore flow rates (20 µl/min to 3 ml/min). The newsoftware features single-point control from an external computerof all detection parameters and programmed methods, as well asmanual and automated control of external events and real-timedata monitoring. The software allows users to easily create, store,and recall operation methods and collected data files directly on aWindows-based personal computer. The Chromachem ELSD is suit-ed for the detection of a wide range of weakly chromophoric com-pounds, including lipids, phospholipids, triglycerides, carbohydrates,polymers, pharmaceuticals, and combinatorial library compounds.

PEPTIDE TIPSThe PerfectPure C-18 Tip is a toolfor desalting, purifying, and concen-trating peptides for mass spectro-metry. The tips’ most significant advantage is the high degree of

reproducibility of the flow-through characteristics, which is achievedwith innovative manufacturing technology and strict quality control.The product specifications are tailored to meet the requirements ofpeptide mass fingerprinting, the identification of protein samples using mass spectrometry analysis. Alternative elution protocols,which may increase peptide recovery, are available for more complex matrices.The tip fits on most common pipet brands.

CELL MIGRATION RESPONSEUsing Guava Express and the Gua-va PCA cell analysis system, re-searchers can quickly and easilystudy cell migration and screen fordrug compounds that affect that

process. The use of Guava Express in an actin polymerization as-say can provide an early measure of chemotaxis, within secondsof exposure to the stimuli. The actin assay can be readily adopt-ed to identify drug compounds that enhance or inhibit chemo-taxis or to study pathways involved in chemotaxis.

PCR WORKSTATIONSThe PCR Workstation offers micro-scope compatibility, accommodat-ing both single and dual ocular microscopes. The workstation isdesigned to protect against poly-

merase chain reaction (PCR) contamination in two ways. Ultra-violet (UV) irradiation of the working area prior to use blocksreplication of contaminating DNA sequences by causing adjacentpyrimidines to undergo dimerization. The workstation protects

against cross contamination or airborne contamination by limit-ing exposure of the experimental set-up to the open lab environ-ment. A small oval cut-out is removed from the tempered glassfacia to accommodate the oculars. The opening is then coveredwith silicone, which permits transmittance of only negligibleamounts of both UVA and UVB, and is cut to securely wrap theoculars. The silicone also protects the sterility of the microscopestage area by preventing contaminating DNA sequences fromfalling through the ocular cut-out.

DIGITAL CAMERAS FORMICROBIOLOGYA range of high-resolution, colordigital cameras can be connectedto a ProtoCOL automated colonycounting and zone sizing system as

a cost-effective method to automatically analyze samples via amicroscope or on bioassay plates. The cameras’ color reproduc-tion and high resolving power make them suitable for accuratelyimaging colonies, zones, or plaques. The cameras come equippedwith the latest Fire-Wire technology toallow microbiologiststo see real-time im-ages and save timewhen analyzing sam-ples. The cameras en-able users to read awide range of mediaand plates types.

RNA ISOLATIONThe RNAqueous-4PCR Kit isolatesRNA without genomic DNA con-tamination from samples as smallas 1 mg or 100 cells. The rapid, phe-nol-free isolation procedure yields

total RNA that is treated with deoxyribonuclease (DNase) in anoptimized reaction buffer. The DNase and divalent cations are re-moved with a novel, nontoxic DNase Removal Reagent providingtotal RNA ready for reverse transcription-polymerase chain reac-tion analysis (RT-PRC). It is designed for both traditional end-point RT-PCR analysis and real-time PCR applications.

LITERATUREStoelting 2004–2005 is a physiolo-gy research instruments catalog forneuroscience, physiology, and gen-eral laboratory equipment. This112-page, illustrated catalog lists

hundreds of products from surgical needs to stereotaxic instru-ments, including more than 50 new products.

Newly offered instrumentation, apparatus, and laboratory materials of interest to re-searchers in all disciplines in academic, industrial, and government organizations arefeatured in this space. Emphasis is given to purpose, chief characteristics, and availabil-ity of products and materials. Endorsement by Science or AAAS of any products ormaterials mentioned is not implied. Additional information may be obtained from themanufacturer or supplier by visiting www.scienceproductlink.org on the Web, whereyou can request that the information be sent to you by e-mail, fax, mail, or telephone.

N E W P R O D U C T SESA

For more information

+44 1844-239381

www.esainc.com

www.scienceproductlink.org

Stoelting


630-860-9700

www.stoeltingco.com


Ambion


800-888-8804

www.ambion.com


Synbiosis


+44 (0) 1223 727125

www.synbiosis.com


C.B.S. Scientific


800-243-4959

www.cbsscientific.com


Eppendorf


+49 40-5 38 01-0

www.eppendorf.com


Guava Technologies


866-448-2827

www.guavatechnologies.com


science.magazine.5684.2004-07-30

Documents

osedax worms

male worms

worms burrow

surface energy

hydrothermal vent worms

gray whale

genus osedax

science vol